feat(backend/copilot): local-LLM AutoPilot for the no-API-key install#12993
feat(backend/copilot): local-LLM AutoPilot for the no-API-key install#12993ntindle wants to merge 16 commits into
Conversation
github-project-automation
Bot
moved this to 🆕 Needs initial review
in AutoGPT development kanban
github-actions
Bot
added
documentation
|
Note Reviews pausedIt looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the Use the following commands to manage reviews:
Use the checkboxes below for quick actions:
WalkthroughAdds a "local" transport to route AutoPilot chat to self-hosted OpenAI‑compatible LLMs (e.g., Ollama). Introduces TransportProfile/TransportName, transport-driven defaults/validators, per‑turn thinking_available kill‑switch, SDK guards, local context/timeout wiring, installer support to bootstrap Ollama, tests, and docs. ChangesLocal transport + routing, SDK guards, installer, docs, tests
Sequence DiagramsequenceDiagram
autonumber
actor User
participant Backend as "AutoPilot Backend"
participant Config as "ChatConfig / TransportProfile"
participant LocalLLM as "Ollama / Local LLM"
participant SDK as "SDK (Langfuse/OpenAI client)"
User->>Backend: POST /copilot (chat request)
Backend->>Config: resolve transport & thinking_available
alt transport == local
Backend->>LocalLLM: Send request (base_url, api_key, num_ctx, timeout)
LocalLLM-->>Backend: Response
Backend-->>User: Return chat response
else transport supports SDK
Backend->>SDK: build_sdk_env() (guarded by transport.supports_sdk)
SDK-->>Backend: SDK response (extended thinking path)
Backend-->>User: Return chat response
end
Estimated code review effort🎯 4 (Complex) | ⏱️ ~60 minutes Possibly related PRs
Suggested labels
Suggested reviewers
Poem
🚥 Pre-merge checks | ✅ 4 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (4 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
🔍 PR Overlap DetectionThis check compares your PR against all other open PRs targeting the same branch to detect potential merge conflicts early. 🔴 Merge Conflicts DetectedThe following PRs have been tested and will have merge conflicts if merged after this PR. Consider coordinating with the authors.
🟢 Low Risk — File Overlap OnlyThese PRs touch the same files but different sections (click to expand)
Summary: 3 conflict(s), 0 medium risk, 14 low risk (out of 17 PRs with file overlap) Auto-generated on push. Ignores: |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 4
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@autogpt_platform/backend/.env.default`:
- Around line 78-90: Reorder the three AutoPilot env keys in .env.default so
they follow dotenv-linter's expected order: place CHAT_API_KEY first, then
CHAT_BASE_URL, then CHAT_USE_LOCAL (preserve their existing values and
surrounding comment block). Update the block containing CHAT_USE_LOCAL,
CHAT_BASE_URL, CHAT_API_KEY to the new sequence (referencing the exact variable
names CHAT_API_KEY, CHAT_BASE_URL, CHAT_USE_LOCAL) so the linter warning is
resolved.
In `@autogpt_platform/backend/backend/copilot/config.py`:
- Around line 447-494: When use_local is True the config currently allows
falling back to other base URLs; add a post-parse guard (a Pydantic
`@root_validator` or `@validator` with pre=False) on the same Pydantic config model
that defines use_local/local_num_ctx/local_request_timeout_s to raise a clear
ValueError if the OpenAI-compatible client base URL or API key are missing
(check the config fields that map to CHAT_BASE_URL and CHAT_API_KEY), ensuring
misconfiguration is caught at startup; replicate the same validator logic for
the other related config models referenced around the other ranges (the blocks
noted at 649-662 and 698-719).
In `@autogpt_platform/installer/setup-autogpt.sh`:
- Around line 185-187: The current sed call that removes from the marker to EOF
(the '/# === Local-LLM AutoPilot wiring/,$d' invocation) is unsafe and can
delete unrelated user config; change the logic to remove only the bounded block
between a start and end marker (e.g., '# === Local-LLM AutoPilot wiring START'
and '# === Local-LLM AutoPilot wiring END') so reruns are idempotent. Update the
grep/sed sequence that references '# === Local-LLM AutoPilot wiring' to look for
the bounded start marker and delete only the region between start and end
markers (use sed address range deletion like '/START/,/END/d' or equivalent),
and ensure you add the matching end marker when writing the block so future runs
can target the block precisely.
- Around line 132-138: When OLLAMA_HOST_URL is provided the script returns after
checking /api/version and never verifies or pulls the configured OLLAMA_MODEL,
which causes runtime "model not found" errors; update the block guarded by if [
-n "$OLLAMA_HOST_URL" ] (the code that prints "Using existing Ollama at
$OLLAMA_HOST_URL" and currently returns) to call the remote Ollama model
endpoints to confirm the value of OLLAMA_MODEL exists and, if missing, trigger
the appropriate remote model pull (or surface a clear error) before returning —
mirror the local-host logic used elsewhere (the same check referenced around
lines 163-165) so remote hosts get the same model validation/pull behavior.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: 81741ba4-8c91-4dff-aab4-9740ce4426f2
📒 Files selected for processing (12)
autogpt_platform/backend/.env.defaultautogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/config.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/service.pyautogpt_platform/installer/setup-autogpt.shdocs/platform/SUMMARY.mddocs/platform/copilot-local-llm.md
📜 Review details
⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (13)
- GitHub Check: check API types
- GitHub Check: Cursor Bugbot
- GitHub Check: test (3.13)
- GitHub Check: type-check (3.11)
- GitHub Check: type-check (3.13)
- GitHub Check: lint
- GitHub Check: test (3.11)
- GitHub Check: test (3.12)
- GitHub Check: type-check (3.12)
- GitHub Check: end-to-end tests
- GitHub Check: Check PR Status
- GitHub Check: Analyze (typescript)
- GitHub Check: Analyze (python)
🧰 Additional context used
📓 Path-based instructions (5)
autogpt_platform/backend/.env*
📄 CodeRabbit inference engine (autogpt_platform/AGENTS.md)
Backend environment configuration:
backend/.env.defaultprovides defaults (tracked in git),backend/.envprovides user overrides (gitignored)
Files:
autogpt_platform/backend/.env.default
autogpt_platform/**/.env*
📄 CodeRabbit inference engine (autogpt_platform/AGENTS.md)
Platform environment configuration:
.env.defaultprovides Supabase/shared defaults (tracked in git),.envprovides user overrides (gitignored)
Files:
autogpt_platform/backend/.env.default
autogpt_platform/backend/**/*.py
📄 CodeRabbit inference engine (.github/copilot-instructions.md)
autogpt_platform/backend/**/*.py: Use Python 3.11 (required; managed by Poetry via pyproject.toml) for backend development
Always run 'poetry run format' (Black + isort) before linting in backend development
Always run 'poetry run lint' (ruff) after formatting in backend development
autogpt_platform/backend/**/*.py: Usepoetry run ...command for executing Python package dependencies
Use top-level imports only — avoid local/inner imports except for lazy imports of heavy optional dependencies likeopenpyxl
Use absolute imports withfrom backend.module import ...for cross-package imports; single-dot relative imports are acceptable for sibling modules within the same package; avoid double-dot relative imports
Do not use duck typing — avoidhasattr/getattr/isinstancefor type dispatch; use typed interfaces/unions/protocols instead
Use Pydantic models over dataclass/namedtuple/dict for structured data
Do not use linter suppressors — no# type: ignore,# noqa,# pyright: ignore; fix the type/code instead
Prefer list comprehensions over manual loop-and-append patterns
Use early return with guard clauses first to avoid deep nesting
Use%sfor deferred interpolation indebuglog statements for efficiency; use f-strings elsewhere for readability (e.g.,logger.debug("Processing %s items", count)vslogger.info(f"Processing {count} items"))
Sanitize error paths by usingos.path.basename()in error messages to avoid leaking directory structure
Be aware of TOCTOU (Time-Of-Check-Time-Of-Use) issues — avoid check-then-act patterns for file access and credit charging
Usetransaction=Truefor Redis pipelines to ensure atomicity on multi-step operations
Usemax(0, value)guards for computed values that should never be negative
Keep files under ~300 lines; if a file grows beyond this, split by responsibility (extract helpers, models, or a sub-module into a new file)
Keep functions under ~40 lines; extract named helpers when a function grows longer
...
Files:
autogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/config.py
autogpt_platform/{backend,autogpt_libs}/**/*.py
📄 CodeRabbit inference engine (AGENTS.md)
Format Python code with
poetry run format
Files:
autogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/config.py
autogpt_platform/backend/**/*_test.py
📄 CodeRabbit inference engine (autogpt_platform/backend/AGENTS.md)
autogpt_platform/backend/**/*_test.py: Use pytest with snapshot testing for API responses
Colocate test files with source files using*_test.pynaming convention
Mock at boundaries — mock where the symbol is used, not where it's defined; after refactoring, update mock targets to match new module paths
UseAsyncMockfromunittest.mockfor async functions in tests
When writing tests, use Test-Driven Development (TDD): write failing tests marked with@pytest.mark.xfailbefore implementation, then remove the marker once the implementation is complete
When creating snapshots in tests, usepoetry run pytest path/to/test.py --snapshot-update; always review snapshot changes withgit diffbefore committing
Files:
autogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/config_test.py
🧠 Learnings (9)
📚 Learning: 2026-02-26T17:02:22.448Z
Learnt from: Pwuts
Repo: Significant-Gravitas/AutoGPT PR: 12211
File: .pre-commit-config.yaml:160-179
Timestamp: 2026-02-26T17:02:22.448Z
Learning: Keep the pre-commit hook pattern broad for autogpt_platform/backend to ensure OpenAPI schema changes are captured. Do not narrow to backend/api/ alone, since the generated schema depends on Pydantic models across multiple directories (backend/data/, backend/blocks/, backend/copilot/, backend/integrations/, backend/util/). Narrowing could miss schema changes and cause frontend type desynchronization.
Applied to files:
autogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/config.py
📚 Learning: 2026-03-04T08:04:35.881Z
Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12273
File: autogpt_platform/backend/backend/copilot/tools/workspace_files.py:216-220
Timestamp: 2026-03-04T08:04:35.881Z
Learning: In the AutoGPT Copilot backend, ensure that SVG images are not treated as vision image types by excluding 'image/svg+xml' from INLINEABLE_MIME_TYPES and MULTIMODAL_TYPES in tool_adapter.py; the Claude API supports PNG, JPEG, GIF, and WebP for vision. SVGs (XML text) should be handled via the text path instead, not the vision path.
Applied to files:
autogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/config.py
📚 Learning: 2026-04-01T04:17:41.600Z
Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12632
File: autogpt_platform/backend/backend/copilot/tools/workspace_files.py:0-0
Timestamp: 2026-04-01T04:17:41.600Z
Learning: When reviewing AutoGPT Copilot tool implementations, accept that `readOnlyHint=True` (provided via `ToolAnnotations`) may be applied unconditionally to *all* tools—even tools that have side effects (e.g., `bash_exec`, `write_workspace_file`, or other write/save operations). Do **not** flag these tools for having `readOnlyHint=True`; this is intentional to enable fully-parallel dispatch by the Anthropic SDK/CLI and has been E2E validated. Only flag `readOnlyHint` issues if they conflict with the established `ToolAnnotations` behavior (e.g., missing/incorrect propagation relative to the intended annotation mechanism).
Applied to files:
autogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/config.py
📚 Learning: 2026-03-05T15:42:08.207Z
Learnt from: ntindle
Repo: Significant-Gravitas/AutoGPT PR: 12297
File: .claude/skills/backend-check/SKILL.md:14-16
Timestamp: 2026-03-05T15:42:08.207Z
Learning: In Python files under autogpt_platform/backend (recursively), rely on poetry run format to perform formatting (Black + isort) and linting (ruff). Do not run poetry run lint as a separate step after poetry run format, since format already includes linting checks.
Applied to files:
autogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/config.py
📚 Learning: 2026-03-16T16:35:40.236Z
Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12440
File: autogpt_platform/backend/backend/api/features/workflow_import.py:54-63
Timestamp: 2026-03-16T16:35:40.236Z
Learning: Avoid using the word 'competitor' in public-facing identifiers and text. Use neutral naming for API paths, model names, function names, and UI text. Examples: rename 'CompetitorFormat' to 'SourcePlatform', 'convert_competitor_workflow' to 'convert_workflow', '/competitor-workflow' to '/workflow'. Apply this guideline to files under autogpt_platform/backend and autogpt_platform/frontend.
Applied to files:
autogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/config.py
📚 Learning: 2026-03-31T15:37:38.626Z
Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12623
File: autogpt_platform/backend/backend/copilot/tools/agent_generator/fixer.py:37-47
Timestamp: 2026-03-31T15:37:38.626Z
Learning: When validating/constructing Anthropic API model IDs in Significant-Gravitas/AutoGPT, allow the hyphen-separated Claude Opus 4.6 model ID `claude-opus-4-6` (it corresponds to `LlmModel.CLAUDE_4_6_OPUS` in `autogpt_platform/backend/backend/blocks/llm.py`). Do NOT require the dot-separated form in Anthropic contexts. Only OpenRouter routing variants should use the dot separator (e.g., `anthropic/claude-opus-4.6`); `claude-opus-4-6` should be treated as correct when passed to Anthropic, and flagged only if it’s used in the OpenRouter path where the dot form is expected.
Applied to files:
autogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/config.py
📚 Learning: 2026-04-15T02:43:36.890Z
Learnt from: ntindle
Repo: Significant-Gravitas/AutoGPT PR: 12780
File: autogpt_platform/backend/backend/copilot/tools/workspace_files.py:0-0
Timestamp: 2026-04-15T02:43:36.890Z
Learning: When reviewing Python exception handlers, do not flag `isinstance(e, X)` checks as dead/unreachable if the caught exception `X` is a subclass of the exception type being handled. For example, if `X` (e.g., `VirusScanError`) inherits from `ValueError` (directly or via an intermediate class) and it can be raised within an `except ValueError:` block, then `isinstance(e, X)` inside that handler is reachable and should not be treated as dead code.
Applied to files:
autogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/config.py
📚 Learning: 2026-04-22T11:46:04.431Z
Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12881
File: autogpt_platform/backend/backend/copilot/config.py:0-0
Timestamp: 2026-04-22T11:46:04.431Z
Learning: Do not flag the Claude Sonnet 4.6 model ID as incorrect when it uses the project’s established hyphenated convention: `anthropic/claude-sonnet-4-6`. This hyphen form is the intentional, production convention and should be treated as valid (including in files like llm.py, blocks tests, reasoning.py, `_is_anthropic_model` tests, and config defaults). Note that OpenRouter also accepts the dot variant `anthropic/claude-sonnet-4.6`, so either form may be tolerated, but `anthropic/claude-sonnet-4-6` should be considered the standard to match project usage.
Applied to files:
autogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/config.py
📚 Learning: 2026-04-22T11:46:12.892Z
Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12881
File: autogpt_platform/backend/backend/copilot/baseline/service.py:322-332
Timestamp: 2026-04-22T11:46:12.892Z
Learning: In this codebase (Significant-Gravitas/AutoGPT), OpenRouter-routed Anthropic model IDs should use the hyphen-separated convention (e.g., `anthropic/claude-sonnet-4-6`, `anthropic/claude-opus-4-6`). Although OpenRouter may accept both hyphen and dot variants, treat the hyphen-separated form as the intended, correct codebase-wide convention and do not flag it as an error. Only flag the dot-separated variant (e.g., `anthropic/claude-sonnet-4.6`) as incorrect when reviewing/validating model ID strings for OpenRouter-routed Anthropic models.
Applied to files:
autogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/config.py
🪛 dotenv-linter (4.0.0)
autogpt_platform/backend/.env.default
[warning] 89-89: [UnorderedKey] The CHAT_BASE_URL key should go before the CHAT_USE_LOCAL key
(UnorderedKey)
[warning] 90-90: [UnorderedKey] The CHAT_API_KEY key should go before the CHAT_BASE_URL key
(UnorderedKey)
🪛 LanguageTool
docs/platform/copilot-local-llm.md
[grammar] ~221-~221: Ensure spelling is correct
Context: ...T_API_KEY` so a stray cloud key set for graphiti / embedders doesn't silently bind to your ...
(QB_NEW_EN_ORTHOGRAPHY_ERROR_IDS_1)
🔇 Additional comments (10)
docs/platform/SUMMARY.md (1)
29-29: TOC entry looks good.The new link is clear and matches the added local-LLM documentation.
autogpt_platform/backend/backend/copilot/sdk/env.py (1)
51-62: Fail-fast guard looks correct.Rejecting SDK env construction when the active transport cannot run the SDK is the right protection here.
autogpt_platform/backend/backend/copilot/service.py (1)
68-76: Local timeout wiring looks good.Extending the OpenAI client timeout only for
localkeeps the default behavior unchanged for cloud transports.autogpt_platform/backend/backend/copilot/baseline/service.py (1)
680-690: Localnum_ctxoverride is applied in the right place.Scoping it to
config.transport.name == "local"keeps the OpenAI/OpenRouter payload unchanged.autogpt_platform/backend/backend/copilot/executor/processor_test.py (1)
137-180: Good regression coverage for the new kill-switch.This locks in the baseline downgrade, the warning path, and the legacy default behavior.
autogpt_platform/backend/backend/copilot/sdk/env_test.py (1)
482-499: Nice regression test for the SDK env guard.This confirms the builder fails fast under the local transport instead of emitting a broken subprocess env.
autogpt_platform/backend/backend/copilot/executor/processor.py (2)
145-173: The newthinking_availablegate is doing the right thing.Short-circuiting to baseline before the mode/flag checks is the correct behavior for SDK-incompatible transports.
513-519: Wiring the transport flag through the call site is correct.Passing
config.thinking_availablehere makes the new routing guard effective during turn execution.docs/platform/copilot-local-llm.md (1)
1-232: Well-structured local-LLM runbook.This doc is clear, operationally practical, and maps cleanly to the config/installer behavior introduced in this PR.
autogpt_platform/backend/backend/copilot/config_test.py (1)
19-19: Strong coverage additions for transport semantics and local safety behavior.The new tests pin transport precedence, fallback policy, and local aux-model derivation well, reducing regression risk in config routing.
Also applies to: 296-526
Codecov Report❌ Patch coverage is Additional details and impacted files@@ Coverage Diff @@
## dev #12993 +/- ##
==========================================
- Coverage 71.18% 70.77% -0.42%
==========================================
Files 2203 2192 -11
Lines 166435 166565 +130
Branches 16966 16973 +7
==========================================
- Hits 118483 117888 -595
- Misses 44460 45283 +823
+ Partials 3492 3394 -98
Flags with carried forward coverage won't be shown. Click here to find out more.
🚀 New features to boost your workflow:
|
ntindle
force-pushed
the
feat/copilot-local-ollama-transport
branch
from
b25340d to
29b9dfd
Compare
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 1
🧹 Nitpick comments (2)
autogpt_platform/backend/backend/copilot/executor/processor.py (1)
298-302: ⚡ Quick winAvoid introducing a new inline
# type: ignorein the CLI prewarm path.This changed block now depends on a private
claude-agent-sdkAPI and suppresses the type error at the call site. Please move this behind a small typed helper or otherwise fix the signature mismatch directly instead of adding a new backend linter suppressor.As per coding guidelines, "Do not use linter suppressors — no
# type: ignore,# noqa,# pyright: ignore; fix the type/code instead."🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@autogpt_platform/backend/backend/copilot/executor/processor.py` around lines 298 - 302, The new inline "# type: ignore" should be removed and the private API call to SubprocessCLITransport._find_bundled_cli should be wrapped by a small typed helper that resolves the signature mismatch; create a helper function (e.g., def _get_bundled_cli_path(cli_arg: Optional[str] = None) -> str) that imports SubprocessCLITransport and calls SubprocessCLITransport._find_bundled_cli(cli_arg) while handling/typing any None/default parameter correctly (or using typing.cast when necessary), then replace the inline call in processor.py with a call to that helper instead of suppressing type checks.autogpt_platform/backend/backend/copilot/config.py (1)
22-57: ⚡ Quick winPrefer a Pydantic model for
TransportProfile.This new transport descriptor is structured backend config data, but it's introduced as a
dataclass. Converting it to a small frozen Pydantic model would keep this module aligned with the rest of the backend config surface and the repo standard.As per coding guidelines, "Use Pydantic models over dataclass/namedtuple/dict for structured data."
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@autogpt_platform/installer/setup-autogpt.sh`:
- Around line 141-142: The conditional that checks for the remote model uses
plain grep which treats regex metacharacters in $OLLAMA_MODEL as special; change
the lookup to use fixed-string matching (grep -F or --fixed-strings) so the
literal pattern "\"name\":\"$OLLAMA_MODEL\"" is searched for instead of a regex.
Update the conditional that calls curl and pipes to grep (the block referencing
OLLAMA_HOST_URL and OLLAMA_MODEL) to use fixed-string grep with -q so model
names like "llama3.1:8b-instruct-q4_K_M" are matched literally.
---
Nitpick comments:
In `@autogpt_platform/backend/backend/copilot/executor/processor.py`:
- Around line 298-302: The new inline "# type: ignore" should be removed and the
private API call to SubprocessCLITransport._find_bundled_cli should be wrapped
by a small typed helper that resolves the signature mismatch; create a helper
function (e.g., def _get_bundled_cli_path(cli_arg: Optional[str] = None) -> str)
that imports SubprocessCLITransport and calls
SubprocessCLITransport._find_bundled_cli(cli_arg) while handling/typing any
None/default parameter correctly (or using typing.cast when necessary), then
replace the inline call in processor.py with a call to that helper instead of
suppressing type checks.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: c9b2665c-4d02-4c88-9ce2-be61967888fa
📒 Files selected for processing (12)
autogpt_platform/backend/.env.defaultautogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/config.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/service.pyautogpt_platform/installer/setup-autogpt.shdocs/platform/SUMMARY.mddocs/platform/copilot-local-llm.md
✅ Files skipped from review due to trivial changes (2)
- docs/platform/SUMMARY.md
- docs/platform/copilot-local-llm.md
📜 Review details
⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (10)
- GitHub Check: check API types
- GitHub Check: Cursor Bugbot
- GitHub Check: end-to-end tests
- GitHub Check: test (3.13)
- GitHub Check: test (3.11)
- GitHub Check: type-check (3.12)
- GitHub Check: test (3.12)
- GitHub Check: Analyze (python)
- GitHub Check: Check PR Status
- GitHub Check: Analyze (typescript)
🧰 Additional context used
📓 Path-based instructions (5)
autogpt_platform/backend/**/*.py
📄 CodeRabbit inference engine (.github/copilot-instructions.md)
autogpt_platform/backend/**/*.py: Use Python 3.11 (required; managed by Poetry via pyproject.toml) for backend development
Always run 'poetry run format' (Black + isort) before linting in backend development
Always run 'poetry run lint' (ruff) after formatting in backend development
autogpt_platform/backend/**/*.py: Usepoetry run ...command for executing Python package dependencies
Use top-level imports only — avoid local/inner imports except for lazy imports of heavy optional dependencies likeopenpyxl
Use absolute imports withfrom backend.module import ...for cross-package imports; single-dot relative imports are acceptable for sibling modules within the same package; avoid double-dot relative imports
Do not use duck typing — avoidhasattr/getattr/isinstancefor type dispatch; use typed interfaces/unions/protocols instead
Use Pydantic models over dataclass/namedtuple/dict for structured data
Do not use linter suppressors — no# type: ignore,# noqa,# pyright: ignore; fix the type/code instead
Prefer list comprehensions over manual loop-and-append patterns
Use early return with guard clauses first to avoid deep nesting
Use%sfor deferred interpolation indebuglog statements for efficiency; use f-strings elsewhere for readability (e.g.,logger.debug("Processing %s items", count)vslogger.info(f"Processing {count} items"))
Sanitize error paths by usingos.path.basename()in error messages to avoid leaking directory structure
Be aware of TOCTOU (Time-Of-Check-Time-Of-Use) issues — avoid check-then-act patterns for file access and credit charging
Usetransaction=Truefor Redis pipelines to ensure atomicity on multi-step operations
Usemax(0, value)guards for computed values that should never be negative
Keep files under ~300 lines; if a file grows beyond this, split by responsibility (extract helpers, models, or a sub-module into a new file)
Keep functions under ~40 lines; extract named helpers when a function grows longer
...
Files:
autogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/config.py
autogpt_platform/{backend,autogpt_libs}/**/*.py
📄 CodeRabbit inference engine (AGENTS.md)
Format Python code with
poetry run format
Files:
autogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/config.py
autogpt_platform/backend/**/*_test.py
📄 CodeRabbit inference engine (autogpt_platform/backend/AGENTS.md)
autogpt_platform/backend/**/*_test.py: Use pytest with snapshot testing for API responses
Colocate test files with source files using*_test.pynaming convention
Mock at boundaries — mock where the symbol is used, not where it's defined; after refactoring, update mock targets to match new module paths
UseAsyncMockfromunittest.mockfor async functions in tests
When writing tests, use Test-Driven Development (TDD): write failing tests marked with@pytest.mark.xfailbefore implementation, then remove the marker once the implementation is complete
When creating snapshots in tests, usepoetry run pytest path/to/test.py --snapshot-update; always review snapshot changes withgit diffbefore committing
Files:
autogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/config_test.py
autogpt_platform/backend/.env*
📄 CodeRabbit inference engine (autogpt_platform/AGENTS.md)
Backend environment configuration:
backend/.env.defaultprovides defaults (tracked in git),backend/.envprovides user overrides (gitignored)
Files:
autogpt_platform/backend/.env.default
autogpt_platform/**/.env*
📄 CodeRabbit inference engine (autogpt_platform/AGENTS.md)
Platform environment configuration:
.env.defaultprovides Supabase/shared defaults (tracked in git),.envprovides user overrides (gitignored)
Files:
autogpt_platform/backend/.env.default
🧠 Learnings (9)
📚 Learning: 2026-02-26T17:02:22.448Z
Learnt from: Pwuts
Repo: Significant-Gravitas/AutoGPT PR: 12211
File: .pre-commit-config.yaml:160-179
Timestamp: 2026-02-26T17:02:22.448Z
Learning: Keep the pre-commit hook pattern broad for autogpt_platform/backend to ensure OpenAPI schema changes are captured. Do not narrow to backend/api/ alone, since the generated schema depends on Pydantic models across multiple directories (backend/data/, backend/blocks/, backend/copilot/, backend/integrations/, backend/util/). Narrowing could miss schema changes and cause frontend type desynchronization.
Applied to files:
autogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/config.py
📚 Learning: 2026-03-04T08:04:35.881Z
Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12273
File: autogpt_platform/backend/backend/copilot/tools/workspace_files.py:216-220
Timestamp: 2026-03-04T08:04:35.881Z
Learning: In the AutoGPT Copilot backend, ensure that SVG images are not treated as vision image types by excluding 'image/svg+xml' from INLINEABLE_MIME_TYPES and MULTIMODAL_TYPES in tool_adapter.py; the Claude API supports PNG, JPEG, GIF, and WebP for vision. SVGs (XML text) should be handled via the text path instead, not the vision path.
Applied to files:
autogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/config.py
📚 Learning: 2026-04-01T04:17:41.600Z
Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12632
File: autogpt_platform/backend/backend/copilot/tools/workspace_files.py:0-0
Timestamp: 2026-04-01T04:17:41.600Z
Learning: When reviewing AutoGPT Copilot tool implementations, accept that `readOnlyHint=True` (provided via `ToolAnnotations`) may be applied unconditionally to *all* tools—even tools that have side effects (e.g., `bash_exec`, `write_workspace_file`, or other write/save operations). Do **not** flag these tools for having `readOnlyHint=True`; this is intentional to enable fully-parallel dispatch by the Anthropic SDK/CLI and has been E2E validated. Only flag `readOnlyHint` issues if they conflict with the established `ToolAnnotations` behavior (e.g., missing/incorrect propagation relative to the intended annotation mechanism).
Applied to files:
autogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/config.py
📚 Learning: 2026-03-05T15:42:08.207Z
Learnt from: ntindle
Repo: Significant-Gravitas/AutoGPT PR: 12297
File: .claude/skills/backend-check/SKILL.md:14-16
Timestamp: 2026-03-05T15:42:08.207Z
Learning: In Python files under autogpt_platform/backend (recursively), rely on poetry run format to perform formatting (Black + isort) and linting (ruff). Do not run poetry run lint as a separate step after poetry run format, since format already includes linting checks.
Applied to files:
autogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/config.py
📚 Learning: 2026-03-16T16:35:40.236Z
Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12440
File: autogpt_platform/backend/backend/api/features/workflow_import.py:54-63
Timestamp: 2026-03-16T16:35:40.236Z
Learning: Avoid using the word 'competitor' in public-facing identifiers and text. Use neutral naming for API paths, model names, function names, and UI text. Examples: rename 'CompetitorFormat' to 'SourcePlatform', 'convert_competitor_workflow' to 'convert_workflow', '/competitor-workflow' to '/workflow'. Apply this guideline to files under autogpt_platform/backend and autogpt_platform/frontend.
Applied to files:
autogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/config.py
📚 Learning: 2026-03-31T15:37:38.626Z
Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12623
File: autogpt_platform/backend/backend/copilot/tools/agent_generator/fixer.py:37-47
Timestamp: 2026-03-31T15:37:38.626Z
Learning: When validating/constructing Anthropic API model IDs in Significant-Gravitas/AutoGPT, allow the hyphen-separated Claude Opus 4.6 model ID `claude-opus-4-6` (it corresponds to `LlmModel.CLAUDE_4_6_OPUS` in `autogpt_platform/backend/backend/blocks/llm.py`). Do NOT require the dot-separated form in Anthropic contexts. Only OpenRouter routing variants should use the dot separator (e.g., `anthropic/claude-opus-4.6`); `claude-opus-4-6` should be treated as correct when passed to Anthropic, and flagged only if it’s used in the OpenRouter path where the dot form is expected.
Applied to files:
autogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/config.py
📚 Learning: 2026-04-15T02:43:36.890Z
Learnt from: ntindle
Repo: Significant-Gravitas/AutoGPT PR: 12780
File: autogpt_platform/backend/backend/copilot/tools/workspace_files.py:0-0
Timestamp: 2026-04-15T02:43:36.890Z
Learning: When reviewing Python exception handlers, do not flag `isinstance(e, X)` checks as dead/unreachable if the caught exception `X` is a subclass of the exception type being handled. For example, if `X` (e.g., `VirusScanError`) inherits from `ValueError` (directly or via an intermediate class) and it can be raised within an `except ValueError:` block, then `isinstance(e, X)` inside that handler is reachable and should not be treated as dead code.
Applied to files:
autogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/config.py
📚 Learning: 2026-04-22T11:46:04.431Z
Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12881
File: autogpt_platform/backend/backend/copilot/config.py:0-0
Timestamp: 2026-04-22T11:46:04.431Z
Learning: Do not flag the Claude Sonnet 4.6 model ID as incorrect when it uses the project’s established hyphenated convention: `anthropic/claude-sonnet-4-6`. This hyphen form is the intentional, production convention and should be treated as valid (including in files like llm.py, blocks tests, reasoning.py, `_is_anthropic_model` tests, and config defaults). Note that OpenRouter also accepts the dot variant `anthropic/claude-sonnet-4.6`, so either form may be tolerated, but `anthropic/claude-sonnet-4-6` should be considered the standard to match project usage.
Applied to files:
autogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/config.py
📚 Learning: 2026-04-22T11:46:12.892Z
Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12881
File: autogpt_platform/backend/backend/copilot/baseline/service.py:322-332
Timestamp: 2026-04-22T11:46:12.892Z
Learning: In this codebase (Significant-Gravitas/AutoGPT), OpenRouter-routed Anthropic model IDs should use the hyphen-separated convention (e.g., `anthropic/claude-sonnet-4-6`, `anthropic/claude-opus-4-6`). Although OpenRouter may accept both hyphen and dot variants, treat the hyphen-separated form as the intended, correct codebase-wide convention and do not flag it as an error. Only flag the dot-separated variant (e.g., `anthropic/claude-sonnet-4.6`) as incorrect when reviewing/validating model ID strings for OpenRouter-routed Anthropic models.
Applied to files:
autogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/config.py
🔇 Additional comments (5)
autogpt_platform/backend/backend/copilot/executor/processor_test.py (1)
137-179: LGTM — solid coverage of the newthinking_availablekwarg.
test_thinking_unavailable_forces_baselinecorrectly pins the tight invariant that only the explicit"extended_thinking"path emits a warning (vialen(warnings) == 1), so future regressions where"fast"orNoneaccidentally start warning would be caught.test_thinking_available_default_preserves_legacy_behaviourkeeping the kwarg out of the call site explicitly verifies backwards compatibility for the dozens of existing call sites that don't pass it.autogpt_platform/backend/backend/copilot/config_test.py (4)
19-19: Good catch — prevents env leakage from skewing local-transport tests.Without
CHAT_USE_LOCALin_ENV_VARS_TO_CLEAR, a developer running tests withCHAT_USE_LOCAL=trueexported in their shell would see flapping behavior inTestSdkModelVendorCompatibilityandTestOpenrouterActive(sinceeffective_transportwould silently flip to"local"and bypass the vendor validator). This addition keeps the test suite hermetic alongside the rest of the new transport's env vars.
369-381: Strong security-flavored test — pins the no-silent-leak guarantee.Asserting on the
CHAT_API_KEY-missing error (rather than just oncfg.api_key is None) makes this test simultaneously verify both invariants: (1) the local profile'sapi_key_fallback_envs == ()doesn't pick upOPENAI_API_KEY, and (2) the_validate_local_transport_requirementsguard fails fast at boot instead of letting AutoPilot 401 against Ollama on the first turn. This is exactly the kind of regression-pinning that catches "helpful" future changes that re-add OpenAI as a fallback for local.
296-360: Profile-shape pinning is a good single source of truth.
TestTransportProfilelocks in the descriptor table per transport, which means any future change that scattersif use_local/if use_openrouterbranches back through the codebase will surface as a profile-shape mismatch here rather than as scattered behavioral regressions. Thetest_thinking_available_alias_matches_profilecross-check (alias ↔transport.supports_sdk) is the right invariant to keep the executor's per-turn kill-switch aligned with the config's transport descriptor.
487-570: Comprehensive local-transport coverage with the right defenses.
test_local_transport_overrides_subscriptionandtest_local_transport_overrides_openrouterare the right invariants for the precedence rule — an operator opting into local self-hosting must win against inherited cloud credentials in CI/dev envs.test_local_skips_sdk_vendor_validatorcorrectly pairs withtest_default_use_local_is_falseto confirm the validator skip is gated specifically on the local transport rather than weakening the vendor check globally.
ntindle
force-pushed
the
feat/copilot-local-ollama-transport
branch
from
29b9dfd to
3de03fd
Compare
There was a problem hiding this comment.
Actionable comments posted: 1
🧹 Nitpick comments (1)
autogpt_platform/backend/backend/copilot/config.py (1)
4-5: ⚡ Quick winUse a Pydantic model for
TransportProfile.Line 22 introduces a new structured backend config object as a
dataclass, which diverges from the repo rule for structured data in backend Python modules. Keeping transport descriptors in the same model system asChatConfigavoids mixing validation conventions inside the config layer.As per coding guidelines "Use Pydantic models over dataclass/namedtuple/dict for structured data".
Also applies to: 22-58
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@autogpt_platform/backend/backend/copilot/config.py` around lines 4 - 5, TransportProfile is defined as a dataclass which breaks the repo rule to use Pydantic models for structured backend config; replace the dataclass TransportProfile with a Pydantic BaseModel (keep the same field names/types and Literal usages), update any imports to pull BaseModel and Field from pydantic, and ensure validation/serialization behavior matches ChatConfig (use same patterns for Optional/defaults and validators if present) so the config layer consistently uses Pydantic models.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@autogpt_platform/installer/setup-autogpt.sh`:
- Around line 132-147: Normalize OLLAMA_HOST_URL before using it: strip any
trailing slash and an optional trailing "/v1" so that API probes use a clean
root (e.g., normalize into a variable like OLLAMA_ROOT), then use
"${OLLAMA_ROOT%/}/api/..." for /api/* checks and set CHAT_BASE_URL by appending
a single "/v1" to that normalized root (avoid double "/v1" when constructing
CHAT_BASE_URL); update all usages (the curl probes that reference
OLLAMA_HOST_URL, the model pull logic using OLLAMA_MODEL, and the later
CHAT_BASE_URL assignment) to use the normalized variable consistently and add a
small comment noting the normalization.
---
Nitpick comments:
In `@autogpt_platform/backend/backend/copilot/config.py`:
- Around line 4-5: TransportProfile is defined as a dataclass which breaks the
repo rule to use Pydantic models for structured backend config; replace the
dataclass TransportProfile with a Pydantic BaseModel (keep the same field
names/types and Literal usages), update any imports to pull BaseModel and Field
from pydantic, and ensure validation/serialization behavior matches ChatConfig
(use same patterns for Optional/defaults and validators if present) so the
config layer consistently uses Pydantic models.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: 4edbff4f-bfaa-4749-984a-52680cb49a0a
📒 Files selected for processing (12)
autogpt_platform/backend/.env.defaultautogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/config.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/service.pyautogpt_platform/installer/setup-autogpt.shdocs/platform/SUMMARY.mddocs/platform/copilot-local-llm.md
🚧 Files skipped from review as they are similar to previous changes (4)
- autogpt_platform/backend/backend/copilot/config_test.py
- autogpt_platform/backend/backend/copilot/executor/processor_test.py
- docs/platform/copilot-local-llm.md
- autogpt_platform/backend/backend/copilot/service.py
📜 Review details
⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (9)
- GitHub Check: check API types
- GitHub Check: Analyze (typescript)
- GitHub Check: Analyze (python)
- GitHub Check: Cursor Bugbot
- GitHub Check: test (3.11)
- GitHub Check: test (3.12)
- GitHub Check: test (3.13)
- GitHub Check: end-to-end tests
- GitHub Check: Check PR Status
🧰 Additional context used
📓 Path-based instructions (5)
autogpt_platform/backend/**/*.py
📄 CodeRabbit inference engine (.github/copilot-instructions.md)
autogpt_platform/backend/**/*.py: Use Python 3.11 (required; managed by Poetry via pyproject.toml) for backend development
Always run 'poetry run format' (Black + isort) before linting in backend development
Always run 'poetry run lint' (ruff) after formatting in backend development
autogpt_platform/backend/**/*.py: Usepoetry run ...command for executing Python package dependencies
Use top-level imports only — avoid local/inner imports except for lazy imports of heavy optional dependencies likeopenpyxl
Use absolute imports withfrom backend.module import ...for cross-package imports; single-dot relative imports are acceptable for sibling modules within the same package; avoid double-dot relative imports
Do not use duck typing — avoidhasattr/getattr/isinstancefor type dispatch; use typed interfaces/unions/protocols instead
Use Pydantic models over dataclass/namedtuple/dict for structured data
Do not use linter suppressors — no# type: ignore,# noqa,# pyright: ignore; fix the type/code instead
Prefer list comprehensions over manual loop-and-append patterns
Use early return with guard clauses first to avoid deep nesting
Use%sfor deferred interpolation indebuglog statements for efficiency; use f-strings elsewhere for readability (e.g.,logger.debug("Processing %s items", count)vslogger.info(f"Processing {count} items"))
Sanitize error paths by usingos.path.basename()in error messages to avoid leaking directory structure
Be aware of TOCTOU (Time-Of-Check-Time-Of-Use) issues — avoid check-then-act patterns for file access and credit charging
Usetransaction=Truefor Redis pipelines to ensure atomicity on multi-step operations
Usemax(0, value)guards for computed values that should never be negative
Keep files under ~300 lines; if a file grows beyond this, split by responsibility (extract helpers, models, or a sub-module into a new file)
Keep functions under ~40 lines; extract named helpers when a function grows longer
...
Files:
autogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/config.py
autogpt_platform/{backend,autogpt_libs}/**/*.py
📄 CodeRabbit inference engine (AGENTS.md)
Format Python code with
poetry run format
Files:
autogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/config.py
autogpt_platform/backend/**/*_test.py
📄 CodeRabbit inference engine (autogpt_platform/backend/AGENTS.md)
autogpt_platform/backend/**/*_test.py: Use pytest with snapshot testing for API responses
Colocate test files with source files using*_test.pynaming convention
Mock at boundaries — mock where the symbol is used, not where it's defined; after refactoring, update mock targets to match new module paths
UseAsyncMockfromunittest.mockfor async functions in tests
When writing tests, use Test-Driven Development (TDD): write failing tests marked with@pytest.mark.xfailbefore implementation, then remove the marker once the implementation is complete
When creating snapshots in tests, usepoetry run pytest path/to/test.py --snapshot-update; always review snapshot changes withgit diffbefore committing
Files:
autogpt_platform/backend/backend/copilot/sdk/env_test.py
autogpt_platform/backend/.env*
📄 CodeRabbit inference engine (autogpt_platform/AGENTS.md)
Backend environment configuration:
backend/.env.defaultprovides defaults (tracked in git),backend/.envprovides user overrides (gitignored)
Files:
autogpt_platform/backend/.env.default
autogpt_platform/**/.env*
📄 CodeRabbit inference engine (autogpt_platform/AGENTS.md)
Platform environment configuration:
.env.defaultprovides Supabase/shared defaults (tracked in git),.envprovides user overrides (gitignored)
Files:
autogpt_platform/backend/.env.default
🧠 Learnings (9)
📚 Learning: 2026-02-26T17:02:22.448Z
Learnt from: Pwuts
Repo: Significant-Gravitas/AutoGPT PR: 12211
File: .pre-commit-config.yaml:160-179
Timestamp: 2026-02-26T17:02:22.448Z
Learning: Keep the pre-commit hook pattern broad for autogpt_platform/backend to ensure OpenAPI schema changes are captured. Do not narrow to backend/api/ alone, since the generated schema depends on Pydantic models across multiple directories (backend/data/, backend/blocks/, backend/copilot/, backend/integrations/, backend/util/). Narrowing could miss schema changes and cause frontend type desynchronization.
Applied to files:
autogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/config.py
📚 Learning: 2026-03-04T08:04:35.881Z
Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12273
File: autogpt_platform/backend/backend/copilot/tools/workspace_files.py:216-220
Timestamp: 2026-03-04T08:04:35.881Z
Learning: In the AutoGPT Copilot backend, ensure that SVG images are not treated as vision image types by excluding 'image/svg+xml' from INLINEABLE_MIME_TYPES and MULTIMODAL_TYPES in tool_adapter.py; the Claude API supports PNG, JPEG, GIF, and WebP for vision. SVGs (XML text) should be handled via the text path instead, not the vision path.
Applied to files:
autogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/config.py
📚 Learning: 2026-04-01T04:17:41.600Z
Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12632
File: autogpt_platform/backend/backend/copilot/tools/workspace_files.py:0-0
Timestamp: 2026-04-01T04:17:41.600Z
Learning: When reviewing AutoGPT Copilot tool implementations, accept that `readOnlyHint=True` (provided via `ToolAnnotations`) may be applied unconditionally to *all* tools—even tools that have side effects (e.g., `bash_exec`, `write_workspace_file`, or other write/save operations). Do **not** flag these tools for having `readOnlyHint=True`; this is intentional to enable fully-parallel dispatch by the Anthropic SDK/CLI and has been E2E validated. Only flag `readOnlyHint` issues if they conflict with the established `ToolAnnotations` behavior (e.g., missing/incorrect propagation relative to the intended annotation mechanism).
Applied to files:
autogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/config.py
📚 Learning: 2026-03-05T15:42:08.207Z
Learnt from: ntindle
Repo: Significant-Gravitas/AutoGPT PR: 12297
File: .claude/skills/backend-check/SKILL.md:14-16
Timestamp: 2026-03-05T15:42:08.207Z
Learning: In Python files under autogpt_platform/backend (recursively), rely on poetry run format to perform formatting (Black + isort) and linting (ruff). Do not run poetry run lint as a separate step after poetry run format, since format already includes linting checks.
Applied to files:
autogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/config.py
📚 Learning: 2026-03-16T16:35:40.236Z
Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12440
File: autogpt_platform/backend/backend/api/features/workflow_import.py:54-63
Timestamp: 2026-03-16T16:35:40.236Z
Learning: Avoid using the word 'competitor' in public-facing identifiers and text. Use neutral naming for API paths, model names, function names, and UI text. Examples: rename 'CompetitorFormat' to 'SourcePlatform', 'convert_competitor_workflow' to 'convert_workflow', '/competitor-workflow' to '/workflow'. Apply this guideline to files under autogpt_platform/backend and autogpt_platform/frontend.
Applied to files:
autogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/config.py
📚 Learning: 2026-03-31T15:37:38.626Z
Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12623
File: autogpt_platform/backend/backend/copilot/tools/agent_generator/fixer.py:37-47
Timestamp: 2026-03-31T15:37:38.626Z
Learning: When validating/constructing Anthropic API model IDs in Significant-Gravitas/AutoGPT, allow the hyphen-separated Claude Opus 4.6 model ID `claude-opus-4-6` (it corresponds to `LlmModel.CLAUDE_4_6_OPUS` in `autogpt_platform/backend/backend/blocks/llm.py`). Do NOT require the dot-separated form in Anthropic contexts. Only OpenRouter routing variants should use the dot separator (e.g., `anthropic/claude-opus-4.6`); `claude-opus-4-6` should be treated as correct when passed to Anthropic, and flagged only if it’s used in the OpenRouter path where the dot form is expected.
Applied to files:
autogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/config.py
📚 Learning: 2026-04-15T02:43:36.890Z
Learnt from: ntindle
Repo: Significant-Gravitas/AutoGPT PR: 12780
File: autogpt_platform/backend/backend/copilot/tools/workspace_files.py:0-0
Timestamp: 2026-04-15T02:43:36.890Z
Learning: When reviewing Python exception handlers, do not flag `isinstance(e, X)` checks as dead/unreachable if the caught exception `X` is a subclass of the exception type being handled. For example, if `X` (e.g., `VirusScanError`) inherits from `ValueError` (directly or via an intermediate class) and it can be raised within an `except ValueError:` block, then `isinstance(e, X)` inside that handler is reachable and should not be treated as dead code.
Applied to files:
autogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/config.py
📚 Learning: 2026-04-22T11:46:04.431Z
Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12881
File: autogpt_platform/backend/backend/copilot/config.py:0-0
Timestamp: 2026-04-22T11:46:04.431Z
Learning: Do not flag the Claude Sonnet 4.6 model ID as incorrect when it uses the project’s established hyphenated convention: `anthropic/claude-sonnet-4-6`. This hyphen form is the intentional, production convention and should be treated as valid (including in files like llm.py, blocks tests, reasoning.py, `_is_anthropic_model` tests, and config defaults). Note that OpenRouter also accepts the dot variant `anthropic/claude-sonnet-4.6`, so either form may be tolerated, but `anthropic/claude-sonnet-4-6` should be considered the standard to match project usage.
Applied to files:
autogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/config.py
📚 Learning: 2026-04-22T11:46:12.892Z
Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12881
File: autogpt_platform/backend/backend/copilot/baseline/service.py:322-332
Timestamp: 2026-04-22T11:46:12.892Z
Learning: In this codebase (Significant-Gravitas/AutoGPT), OpenRouter-routed Anthropic model IDs should use the hyphen-separated convention (e.g., `anthropic/claude-sonnet-4-6`, `anthropic/claude-opus-4-6`). Although OpenRouter may accept both hyphen and dot variants, treat the hyphen-separated form as the intended, correct codebase-wide convention and do not flag it as an error. Only flag the dot-separated variant (e.g., `anthropic/claude-sonnet-4.6`) as incorrect when reviewing/validating model ID strings for OpenRouter-routed Anthropic models.
Applied to files:
autogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/config.py
ntindle
force-pushed
the
feat/copilot-local-ollama-transport
branch
from
3de03fd to
d6bfe04
Compare
There was a problem hiding this comment.
Actionable comments posted: 2
🧹 Nitpick comments (1)
autogpt_platform/installer/setup-autogpt.sh (1)
142-153: ⚡ Quick winAdd
--max-timeto remote Ollama probes to prevent installer hangs.
curl -sfhas no default timeout — if$OLLAMA_ROOTaccepts the TCP connection but the HTTP response stalls (e.g. misconfigured reverse proxy, packet filter that blackholes after handshake), the installer blocks indefinitely with no output and nohandle_errorpath. A short connect+read budget converts that into the same clear "cannot reach" failure as a refused connection.Pull (line 155-158) should keep its lack of timeout — model downloads legitimately take many minutes.
♻️ Proposed change
- if ! curl -sf "${OLLAMA_ROOT}/api/version" > /dev/null; then + if ! curl -sf --max-time 5 "${OLLAMA_ROOT}/api/version" > /dev/null; then handle_error "Cannot reach Ollama at $OLLAMA_ROOT — is it running and listening on 0.0.0.0?" fi @@ - if ! curl -sf "${OLLAMA_ROOT}/api/tags" \ + if ! curl -sf --max-time 10 "${OLLAMA_ROOT}/api/tags" \ | grep -Fq "\"name\":\"$OLLAMA_MODEL\""; then🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@autogpt_platform/installer/setup-autogpt.sh` around lines 142 - 153, The remote Ollama probes using curl to check "${OLLAMA_ROOT}/api/version" and "${OLLAMA_ROOT}/api/tags" can hang indefinitely; update those two curl invocations to include a short overall timeout (e.g. --max-time 10) so stalled HTTP responses fail fast and trigger handle_error, but do not add a timeout to the model download/pull step (leave the curl that performs the model pull unchanged). Ensure you update the two checks that reference OLLAMA_ROOT and the grep for "\"name\":\"$OLLAMA_MODEL\"" while keeping the model download logic as-is.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@autogpt_platform/backend/backend/copilot/config_test.py`:
- Around line 463-470: Hoist the local import of OPENROUTER_BASE_URL out of the
test function into the module-level imports: add "from backend.util.clients
import OPENROUTER_BASE_URL" to the top import block of this test module and
remove the inner import inside
test_explicit_openrouter_base_url_under_local_raises; this preserves the
existing reference to OPENROUTER_BASE_URL in that test and adheres to the
top-level import guideline.
In `@autogpt_platform/backend/backend/copilot/config.py`:
- Around line 22-89: TransportProfile is currently a dataclass but should be a
Pydantic model; replace the `@dataclass`(frozen=True) usage with class
TransportProfile(BaseModel):, add model_config = ConfigDict(frozen=True) inside
the class, keep the same typed attributes (name: TransportName, supports_sdk:
bool, sdk_model_vendor_constraint: str | None, api_key_fallback_envs: tuple[str,
...], inherit_fast_model_for_aux: bool), and import BaseModel and ConfigDict
from pydantic so the existing _TRANSPORT_PROFILES construction and read-only
access semantics remain unchanged.
---
Nitpick comments:
In `@autogpt_platform/installer/setup-autogpt.sh`:
- Around line 142-153: The remote Ollama probes using curl to check
"${OLLAMA_ROOT}/api/version" and "${OLLAMA_ROOT}/api/tags" can hang
indefinitely; update those two curl invocations to include a short overall
timeout (e.g. --max-time 10) so stalled HTTP responses fail fast and trigger
handle_error, but do not add a timeout to the model download/pull step (leave
the curl that performs the model pull unchanged). Ensure you update the two
checks that reference OLLAMA_ROOT and the grep for "\"name\":\"$OLLAMA_MODEL\""
while keeping the model download logic as-is.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: 28976447-cf9d-4c13-b186-470a7c880864
📒 Files selected for processing (12)
autogpt_platform/backend/.env.defaultautogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/config.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/service.pyautogpt_platform/installer/setup-autogpt.shdocs/platform/SUMMARY.mddocs/platform/copilot-local-llm.md
🚧 Files skipped from review as they are similar to previous changes (1)
- autogpt_platform/backend/backend/copilot/baseline/service.py
📜 Review details
⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (11)
- GitHub Check: check API types
- GitHub Check: Cursor Bugbot
- GitHub Check: end-to-end tests
- GitHub Check: type-check (3.11)
- GitHub Check: test (3.13)
- GitHub Check: test (3.12)
- GitHub Check: test (3.11)
- GitHub Check: type-check (3.12)
- GitHub Check: Analyze (python)
- GitHub Check: Analyze (typescript)
- GitHub Check: Check PR Status
🧰 Additional context used
📓 Path-based instructions (5)
autogpt_platform/backend/**/*.py
📄 CodeRabbit inference engine (.github/copilot-instructions.md)
autogpt_platform/backend/**/*.py: Use Python 3.11 (required; managed by Poetry via pyproject.toml) for backend development
Always run 'poetry run format' (Black + isort) before linting in backend development
Always run 'poetry run lint' (ruff) after formatting in backend development
autogpt_platform/backend/**/*.py: Usepoetry run ...command for executing Python package dependencies
Use top-level imports only — avoid local/inner imports except for lazy imports of heavy optional dependencies likeopenpyxl
Use absolute imports withfrom backend.module import ...for cross-package imports; single-dot relative imports are acceptable for sibling modules within the same package; avoid double-dot relative imports
Do not use duck typing — avoidhasattr/getattr/isinstancefor type dispatch; use typed interfaces/unions/protocols instead
Use Pydantic models over dataclass/namedtuple/dict for structured data
Do not use linter suppressors — no# type: ignore,# noqa,# pyright: ignore; fix the type/code instead
Prefer list comprehensions over manual loop-and-append patterns
Use early return with guard clauses first to avoid deep nesting
Use%sfor deferred interpolation indebuglog statements for efficiency; use f-strings elsewhere for readability (e.g.,logger.debug("Processing %s items", count)vslogger.info(f"Processing {count} items"))
Sanitize error paths by usingos.path.basename()in error messages to avoid leaking directory structure
Be aware of TOCTOU (Time-Of-Check-Time-Of-Use) issues — avoid check-then-act patterns for file access and credit charging
Usetransaction=Truefor Redis pipelines to ensure atomicity on multi-step operations
Usemax(0, value)guards for computed values that should never be negative
Keep files under ~300 lines; if a file grows beyond this, split by responsibility (extract helpers, models, or a sub-module into a new file)
Keep functions under ~40 lines; extract named helpers when a function grows longer
...
Files:
autogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/config.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor.py
autogpt_platform/{backend,autogpt_libs}/**/*.py
📄 CodeRabbit inference engine (AGENTS.md)
Format Python code with
poetry run format
Files:
autogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/config.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor.py
autogpt_platform/backend/**/*_test.py
📄 CodeRabbit inference engine (autogpt_platform/backend/AGENTS.md)
autogpt_platform/backend/**/*_test.py: Use pytest with snapshot testing for API responses
Colocate test files with source files using*_test.pynaming convention
Mock at boundaries — mock where the symbol is used, not where it's defined; after refactoring, update mock targets to match new module paths
UseAsyncMockfromunittest.mockfor async functions in tests
When writing tests, use Test-Driven Development (TDD): write failing tests marked with@pytest.mark.xfailbefore implementation, then remove the marker once the implementation is complete
When creating snapshots in tests, usepoetry run pytest path/to/test.py --snapshot-update; always review snapshot changes withgit diffbefore committing
Files:
autogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/config_test.py
autogpt_platform/backend/.env*
📄 CodeRabbit inference engine (autogpt_platform/AGENTS.md)
Backend environment configuration:
backend/.env.defaultprovides defaults (tracked in git),backend/.envprovides user overrides (gitignored)
Files:
autogpt_platform/backend/.env.default
autogpt_platform/**/.env*
📄 CodeRabbit inference engine (autogpt_platform/AGENTS.md)
Platform environment configuration:
.env.defaultprovides Supabase/shared defaults (tracked in git),.envprovides user overrides (gitignored)
Files:
autogpt_platform/backend/.env.default
🧠 Learnings (9)
📚 Learning: 2026-02-26T17:02:22.448Z
Learnt from: Pwuts
Repo: Significant-Gravitas/AutoGPT PR: 12211
File: .pre-commit-config.yaml:160-179
Timestamp: 2026-02-26T17:02:22.448Z
Learning: Keep the pre-commit hook pattern broad for autogpt_platform/backend to ensure OpenAPI schema changes are captured. Do not narrow to backend/api/ alone, since the generated schema depends on Pydantic models across multiple directories (backend/data/, backend/blocks/, backend/copilot/, backend/integrations/, backend/util/). Narrowing could miss schema changes and cause frontend type desynchronization.
Applied to files:
autogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/config.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor.py
📚 Learning: 2026-03-04T08:04:35.881Z
Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12273
File: autogpt_platform/backend/backend/copilot/tools/workspace_files.py:216-220
Timestamp: 2026-03-04T08:04:35.881Z
Learning: In the AutoGPT Copilot backend, ensure that SVG images are not treated as vision image types by excluding 'image/svg+xml' from INLINEABLE_MIME_TYPES and MULTIMODAL_TYPES in tool_adapter.py; the Claude API supports PNG, JPEG, GIF, and WebP for vision. SVGs (XML text) should be handled via the text path instead, not the vision path.
Applied to files:
autogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/config.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor.py
📚 Learning: 2026-04-01T04:17:41.600Z
Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12632
File: autogpt_platform/backend/backend/copilot/tools/workspace_files.py:0-0
Timestamp: 2026-04-01T04:17:41.600Z
Learning: When reviewing AutoGPT Copilot tool implementations, accept that `readOnlyHint=True` (provided via `ToolAnnotations`) may be applied unconditionally to *all* tools—even tools that have side effects (e.g., `bash_exec`, `write_workspace_file`, or other write/save operations). Do **not** flag these tools for having `readOnlyHint=True`; this is intentional to enable fully-parallel dispatch by the Anthropic SDK/CLI and has been E2E validated. Only flag `readOnlyHint` issues if they conflict with the established `ToolAnnotations` behavior (e.g., missing/incorrect propagation relative to the intended annotation mechanism).
Applied to files:
autogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/config.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor.py
📚 Learning: 2026-03-05T15:42:08.207Z
Learnt from: ntindle
Repo: Significant-Gravitas/AutoGPT PR: 12297
File: .claude/skills/backend-check/SKILL.md:14-16
Timestamp: 2026-03-05T15:42:08.207Z
Learning: In Python files under autogpt_platform/backend (recursively), rely on poetry run format to perform formatting (Black + isort) and linting (ruff). Do not run poetry run lint as a separate step after poetry run format, since format already includes linting checks.
Applied to files:
autogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/config.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor.py
📚 Learning: 2026-03-16T16:35:40.236Z
Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12440
File: autogpt_platform/backend/backend/api/features/workflow_import.py:54-63
Timestamp: 2026-03-16T16:35:40.236Z
Learning: Avoid using the word 'competitor' in public-facing identifiers and text. Use neutral naming for API paths, model names, function names, and UI text. Examples: rename 'CompetitorFormat' to 'SourcePlatform', 'convert_competitor_workflow' to 'convert_workflow', '/competitor-workflow' to '/workflow'. Apply this guideline to files under autogpt_platform/backend and autogpt_platform/frontend.
Applied to files:
autogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/config.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor.py
📚 Learning: 2026-03-31T15:37:38.626Z
Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12623
File: autogpt_platform/backend/backend/copilot/tools/agent_generator/fixer.py:37-47
Timestamp: 2026-03-31T15:37:38.626Z
Learning: When validating/constructing Anthropic API model IDs in Significant-Gravitas/AutoGPT, allow the hyphen-separated Claude Opus 4.6 model ID `claude-opus-4-6` (it corresponds to `LlmModel.CLAUDE_4_6_OPUS` in `autogpt_platform/backend/backend/blocks/llm.py`). Do NOT require the dot-separated form in Anthropic contexts. Only OpenRouter routing variants should use the dot separator (e.g., `anthropic/claude-opus-4.6`); `claude-opus-4-6` should be treated as correct when passed to Anthropic, and flagged only if it’s used in the OpenRouter path where the dot form is expected.
Applied to files:
autogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/config.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor.py
📚 Learning: 2026-04-15T02:43:36.890Z
Learnt from: ntindle
Repo: Significant-Gravitas/AutoGPT PR: 12780
File: autogpt_platform/backend/backend/copilot/tools/workspace_files.py:0-0
Timestamp: 2026-04-15T02:43:36.890Z
Learning: When reviewing Python exception handlers, do not flag `isinstance(e, X)` checks as dead/unreachable if the caught exception `X` is a subclass of the exception type being handled. For example, if `X` (e.g., `VirusScanError`) inherits from `ValueError` (directly or via an intermediate class) and it can be raised within an `except ValueError:` block, then `isinstance(e, X)` inside that handler is reachable and should not be treated as dead code.
Applied to files:
autogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/config.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor.py
📚 Learning: 2026-04-22T11:46:04.431Z
Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12881
File: autogpt_platform/backend/backend/copilot/config.py:0-0
Timestamp: 2026-04-22T11:46:04.431Z
Learning: Do not flag the Claude Sonnet 4.6 model ID as incorrect when it uses the project’s established hyphenated convention: `anthropic/claude-sonnet-4-6`. This hyphen form is the intentional, production convention and should be treated as valid (including in files like llm.py, blocks tests, reasoning.py, `_is_anthropic_model` tests, and config defaults). Note that OpenRouter also accepts the dot variant `anthropic/claude-sonnet-4.6`, so either form may be tolerated, but `anthropic/claude-sonnet-4-6` should be considered the standard to match project usage.
Applied to files:
autogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/config.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor.py
📚 Learning: 2026-04-22T11:46:12.892Z
Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12881
File: autogpt_platform/backend/backend/copilot/baseline/service.py:322-332
Timestamp: 2026-04-22T11:46:12.892Z
Learning: In this codebase (Significant-Gravitas/AutoGPT), OpenRouter-routed Anthropic model IDs should use the hyphen-separated convention (e.g., `anthropic/claude-sonnet-4-6`, `anthropic/claude-opus-4-6`). Although OpenRouter may accept both hyphen and dot variants, treat the hyphen-separated form as the intended, correct codebase-wide convention and do not flag it as an error. Only flag the dot-separated variant (e.g., `anthropic/claude-sonnet-4.6`) as incorrect when reviewing/validating model ID strings for OpenRouter-routed Anthropic models.
Applied to files:
autogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/config.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor.py
🔇 Additional comments (15)
autogpt_platform/backend/.env.default (1)
78-90: LGTM!The AutoPilot self-hosted LLM block is well-documented with clear guidance on
CHAT_BASE_URL, model-name conventions, and the extended-thinking auto-downgrade. Keys are now in alphabetical order to satisfydotenv-linter.docs/platform/SUMMARY.md (1)
29-29: LGTM!The new entry slots correctly into "Using AI Services" alongside the existing Ollama page and points to the new AutoPilot-specific doc.
docs/platform/copilot-local-llm.md (1)
1-232: Comprehensive doc — LGTM.Solid coverage of the local-transport contract, model selection, the
num_ctxOllama gotcha, networking topologies (same-host/LAN/remote), wiring verification, and troubleshooting. The Anthropic-wire-protocol downgrade caveat is called out in the right places (intro, behaviour table, troubleshooting).autogpt_platform/installer/setup-autogpt.sh (1)
221-229: 💤 Low valueBounded-block sed delete — verify marker safety.
The bounded
START_MARKER/END_MARKERrewrite is a real improvement over the previous,$dcliff. One thing worth noting:$START_MARKERcontains parentheses ((added by setup-autogpt.sh --with-ollama)) which are literal insedBRE today, but if anyone later refactors the marker to use BRE-special characters (*,^,$,[,],\,/), the address pattern silently breaks. Since markers are constants you control, this is fine as-is — just keep marker characters in the safe set if they ever change.autogpt_platform/backend/backend/copilot/config.py (2)
698-758: 💤 Low valueLGTM on the local-transport guard + fallback chain.
The validator pair correctly fails fast when
CHAT_USE_LOCAL=trueis paired with theOPENROUTER_BASE_URLdefault or a missingCHAT_API_KEY, and the per-transport fallback table elegantly replaces the old scatteredif config.use_Xbranches. The dependency on definition-order execution ofmodel_validator(mode="after")in Pydantic v2 is documented in the docstring, which is the right call.One small simplification opportunity: in Pydantic v2
BaseSettingsdefaults tovalidate_assignment=False, soself.api_key = v(andself.title_model = ...,self.simulation_model = ...) works the same asobject.__setattr__here without the bypass connotation. Not blocking — current form is correct.
779-827: LGTM — vendor-constraint validator now driven bytransport.sdk_model_vendor_constraint.Skip-path correctly short-circuits for transports without an SDK or without a vendor constraint (
local,openrouter,subscription). The error message names the active transport and the offending field, which is the actionable form an operator needs to debug at boot.autogpt_platform/backend/backend/copilot/sdk/env.py (1)
51-62: LGTM — fast-fail guard at the right layer.Raising before any mode logic runs gives a clear, actionable error (with a pointer back to
executor.processor.resolve_use_sdk_for_mode) instead of constructing an env that would route Anthropic-wire traffic to Ollama. Module-levelconfigmakes this patchable from tests, which the newTestBuildSdkEnvLocalTransportGuardexercises.autogpt_platform/backend/backend/copilot/service.py (1)
65-77: LGTM — scoped timeout override for the local transport only.Conditional kwargs cleanly leave cloud transports on the OpenAI client's default 600 s while giving CPU-bound local backends the headroom they need. The
_clientis module-cached so this evaluates once per process — fine sinceconfig.transport.nameis stable for the lifetime of the worker.autogpt_platform/backend/backend/copilot/sdk/env_test.py (1)
482-501: LGTM — defensive-guard test mocks at the use site.Patching
backend.copilot.sdk.env.config(where the symbol is consumed) rather thanbackend.copilot.config.ChatConfig(where it's defined) is the correct boundary per the project's mocking guideline. The regex match also pins the message wording so future refactors can't silently reduce it to a generic error.autogpt_platform/backend/backend/copilot/config_test.py (3)
19-19: LGTM —CHAT_USE_LOCALcorrectly added to_ENV_VARS_TO_CLEAR.
207-217: LGTM — updated error-message match aligns with the new vendor-constraint wording.
296-575: New transport/local test classes look solid.All transport profile shapes, API-key fallback chains, aux-model inheritance, local requirements validation, and
effective_transport/thinking_availablesemantics are covered with clear assertions. Thecaplog.records[i].messageaccess pattern intest_thinking_unavailable_forces_baseline(line 160) is valid — pytest'sLogCaptureHandlerformats records before appending, makingrecord.messagereliably set.autogpt_platform/backend/backend/copilot/executor/processor.py (2)
145-182:thinking_availablekill-switch cleanly slots in with full backward compatibility.The default
thinking_available: bool = Truemeans every existing call site is unaffected. The early return short-circuits all downstream logic (subscription flag, LaunchDarkly,is_feature_enabled) so the baseline is always forced when the transport can't support the SDK. Warning only fires for the explicitly-requestedextended_thinkingcase, which is the right balance — silent downgrade forfast/Nonebut a visible warning when the operator has explicitly asked for thinking.
513-519: LGTM — call site correctly threadsconfig.thinking_availableinto the routing function.autogpt_platform/backend/backend/copilot/executor/processor_test.py (1)
137-179: Newthinking_availabletests are well-structured and cover the right edge cases.
test_thinking_unavailable_forces_baselinecorrectly validates that all three modes ("fast","extended_thinking",None) are forced to baseline whenthinking_available=False, and that exactly oneWARNINGis emitted — only for the explicitly-requestedextended_thinkingmode. Theis_feature_enabledmock (returningTrue) proves the early return fires before any LaunchDarkly call, providing a stronger guarantee than if the mock returnedFalse.
test_thinking_available_default_preserves_legacy_behaviourcorrectly confirms theTruedefault doesn't break any existing behavior.
ntindle
force-pushed
the
feat/copilot-local-ollama-transport
branch
from
d6bfe04 to
91fa945
Compare
There was a problem hiding this comment.
Actionable comments posted: 1
♻️ Duplicate comments (1)
autogpt_platform/backend/backend/copilot/config.py (1)
706-744:⚠️ Potential issue | 🟠 Major | ⚡ Quick win
CHAT_USE_LOCALcan still piggyback inherited base URLs.This validator only rejects the final OpenRouter default, so
use_local=Truestill passes whenbase_urlwas inherited fromOPENAI_BASE_URLorOPENROUTER_BASE_URLinstead of being set explicitly viaCHAT_BASE_URL. That leaves the local transport vulnerable to silent misrouting whenever those global env vars exist in the process.Suggested fix
`@model_validator`(mode="after") def _validate_local_transport_requirements(self) -> "ChatConfig": @@ if self.transport.name != "local": return self + explicit_chat_base_url = os.getenv("CHAT_BASE_URL") + explicit_base_url = bool( + explicit_chat_base_url or "base_url" in self.model_fields_set + ) + if not explicit_base_url: + raise ValueError( + "CHAT_USE_LOCAL=true requires an explicit CHAT_BASE_URL " + "(an OpenAI-compatible /v1 endpoint, e.g. " + "http://host.docker.internal:11434/v1 for Ollama). " + "Do not rely on OPENAI_BASE_URL / OPENROUTER_BASE_URL fallbacks." + ) if not self.base_url or self.base_url.rstrip("/") == OPENROUTER_BASE_URL.rstrip( "/" ): raise ValueError( "CHAT_USE_LOCAL=true requires an explicit CHAT_BASE_URL "🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@autogpt_platform/backend/backend/copilot/config.py` around lines 706 - 744, The validator _validate_local_transport_requirements currently only rejects the OpenRouter default but allows base_url values inherited from other global defaults; update it to also treat any inherited global defaults as invalid by rejecting when self.base_url is missing or equals ANY of the known default endpoints (not just OPENROUTER_BASE_URL) such as OPENAI_BASE_URL and OPENROUTER_BASE_URL (after normalizing trailing slashes), so when self.transport.name == "local" you raise the same ValueError if self.base_url is falsy or matches OPENROUTER_BASE_URL or OPENAI_BASE_URL; keep the existing api_key check and return self otherwise.
🧹 Nitpick comments (1)
autogpt_platform/backend/backend/copilot/executor/processor.py (1)
169-172: ⚡ Quick winMake the downgrade warning transport-agnostic.
This branch is driven by
thinking_available, but the message hardcodesCHAT_USE_LOCAL=true. If another transport disables SDK support later, this log becomes misleading.Proposed tweak
- "Downgrading mode=extended_thinking to fast: SDK is " - "unavailable under the current transport (CHAT_USE_LOCAL=true)" + "Downgrading mode=extended_thinking to fast: SDK is " + "unavailable under the current transport configuration"🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@autogpt_platform/backend/backend/copilot/executor/processor.py` around lines 169 - 172, The warning logged when downgrading "mode=extended_thinking" is hardcoded to "CHAT_USE_LOCAL=true" and should be transport-agnostic; update the logger.warning call in processor.py (where thinking_available is checked) to mention the actual transport or reason instead of a hardcoded env var: include the variable that represents the current transport or a generic phrase like "current transport does not support SDK" (use the existing thinking_available and any transport/transport_name variable available in the surrounding scope) so the message accurately reflects why extended thinking was disabled.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@autogpt_platform/installer/setup-autogpt.sh`:
- Around line 124-187: The script's bootstrap_ollama function assumes curl
exists but check_prerequisites() doesn't validate it, causing unclear failures;
update check_prerequisites() to include curl (and any other tools used in
bootstrap_ollama like grep/tail if not already checked) and/or add an early
explicit check at the start of bootstrap_ollama to test for curl (use command -v
curl) and call handle_error with a clear message if missing; reference the
bootstrap_ollama function and the check_prerequisites function (and the
installer curl usage around the install.sh download and remote probes noted
around the 205-210 area) so the fix is applied where prerequisites are validated
and before any curl usage.
---
Duplicate comments:
In `@autogpt_platform/backend/backend/copilot/config.py`:
- Around line 706-744: The validator _validate_local_transport_requirements
currently only rejects the OpenRouter default but allows base_url values
inherited from other global defaults; update it to also treat any inherited
global defaults as invalid by rejecting when self.base_url is missing or equals
ANY of the known default endpoints (not just OPENROUTER_BASE_URL) such as
OPENAI_BASE_URL and OPENROUTER_BASE_URL (after normalizing trailing slashes), so
when self.transport.name == "local" you raise the same ValueError if
self.base_url is falsy or matches OPENROUTER_BASE_URL or OPENAI_BASE_URL; keep
the existing api_key check and return self otherwise.
---
Nitpick comments:
In `@autogpt_platform/backend/backend/copilot/executor/processor.py`:
- Around line 169-172: The warning logged when downgrading
"mode=extended_thinking" is hardcoded to "CHAT_USE_LOCAL=true" and should be
transport-agnostic; update the logger.warning call in processor.py (where
thinking_available is checked) to mention the actual transport or reason instead
of a hardcoded env var: include the variable that represents the current
transport or a generic phrase like "current transport does not support SDK" (use
the existing thinking_available and any transport/transport_name variable
available in the surrounding scope) so the message accurately reflects why
extended thinking was disabled.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: a6be9cc9-8337-4cb9-a8d5-e47b64d44e0a
📒 Files selected for processing (12)
autogpt_platform/backend/.env.defaultautogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/config.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/service.pyautogpt_platform/installer/setup-autogpt.shdocs/platform/SUMMARY.mddocs/platform/copilot-local-llm.md
🚧 Files skipped from review as they are similar to previous changes (2)
- autogpt_platform/backend/.env.default
- docs/platform/copilot-local-llm.md
📜 Review details
⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (11)
- GitHub Check: check API types
- GitHub Check: Cursor Bugbot
- GitHub Check: test (3.11)
- GitHub Check: type-check (3.11)
- GitHub Check: test (3.13)
- GitHub Check: type-check (3.13)
- GitHub Check: test (3.12)
- GitHub Check: end-to-end tests
- GitHub Check: Check PR Status
- GitHub Check: Analyze (python)
- GitHub Check: Analyze (typescript)
🧰 Additional context used
📓 Path-based instructions (3)
autogpt_platform/backend/**/*.py
📄 CodeRabbit inference engine (.github/copilot-instructions.md)
autogpt_platform/backend/**/*.py: Use Python 3.11 (required; managed by Poetry via pyproject.toml) for backend development
Always run 'poetry run format' (Black + isort) before linting in backend development
Always run 'poetry run lint' (ruff) after formatting in backend development
autogpt_platform/backend/**/*.py: Usepoetry run ...command for executing Python package dependencies
Use top-level imports only — avoid local/inner imports except for lazy imports of heavy optional dependencies likeopenpyxl
Use absolute imports withfrom backend.module import ...for cross-package imports; single-dot relative imports are acceptable for sibling modules within the same package; avoid double-dot relative imports
Do not use duck typing — avoidhasattr/getattr/isinstancefor type dispatch; use typed interfaces/unions/protocols instead
Use Pydantic models over dataclass/namedtuple/dict for structured data
Do not use linter suppressors — no# type: ignore,# noqa,# pyright: ignore; fix the type/code instead
Prefer list comprehensions over manual loop-and-append patterns
Use early return with guard clauses first to avoid deep nesting
Use%sfor deferred interpolation indebuglog statements for efficiency; use f-strings elsewhere for readability (e.g.,logger.debug("Processing %s items", count)vslogger.info(f"Processing {count} items"))
Sanitize error paths by usingos.path.basename()in error messages to avoid leaking directory structure
Be aware of TOCTOU (Time-Of-Check-Time-Of-Use) issues — avoid check-then-act patterns for file access and credit charging
Usetransaction=Truefor Redis pipelines to ensure atomicity on multi-step operations
Usemax(0, value)guards for computed values that should never be negative
Keep files under ~300 lines; if a file grows beyond this, split by responsibility (extract helpers, models, or a sub-module into a new file)
Keep functions under ~40 lines; extract named helpers when a function grows longer
...
Files:
autogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/config.py
autogpt_platform/{backend,autogpt_libs}/**/*.py
📄 CodeRabbit inference engine (AGENTS.md)
Format Python code with
poetry run format
Files:
autogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/config.py
autogpt_platform/backend/**/*_test.py
📄 CodeRabbit inference engine (autogpt_platform/backend/AGENTS.md)
autogpt_platform/backend/**/*_test.py: Use pytest with snapshot testing for API responses
Colocate test files with source files using*_test.pynaming convention
Mock at boundaries — mock where the symbol is used, not where it's defined; after refactoring, update mock targets to match new module paths
UseAsyncMockfromunittest.mockfor async functions in tests
When writing tests, use Test-Driven Development (TDD): write failing tests marked with@pytest.mark.xfailbefore implementation, then remove the marker once the implementation is complete
When creating snapshots in tests, usepoetry run pytest path/to/test.py --snapshot-update; always review snapshot changes withgit diffbefore committing
Files:
autogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor_test.py
🧠 Learnings (9)
📚 Learning: 2026-02-26T17:02:22.448Z
Learnt from: Pwuts
Repo: Significant-Gravitas/AutoGPT PR: 12211
File: .pre-commit-config.yaml:160-179
Timestamp: 2026-02-26T17:02:22.448Z
Learning: Keep the pre-commit hook pattern broad for autogpt_platform/backend to ensure OpenAPI schema changes are captured. Do not narrow to backend/api/ alone, since the generated schema depends on Pydantic models across multiple directories (backend/data/, backend/blocks/, backend/copilot/, backend/integrations/, backend/util/). Narrowing could miss schema changes and cause frontend type desynchronization.
Applied to files:
autogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/config.py
📚 Learning: 2026-03-04T08:04:35.881Z
Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12273
File: autogpt_platform/backend/backend/copilot/tools/workspace_files.py:216-220
Timestamp: 2026-03-04T08:04:35.881Z
Learning: In the AutoGPT Copilot backend, ensure that SVG images are not treated as vision image types by excluding 'image/svg+xml' from INLINEABLE_MIME_TYPES and MULTIMODAL_TYPES in tool_adapter.py; the Claude API supports PNG, JPEG, GIF, and WebP for vision. SVGs (XML text) should be handled via the text path instead, not the vision path.
Applied to files:
autogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/config.py
📚 Learning: 2026-04-01T04:17:41.600Z
Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12632
File: autogpt_platform/backend/backend/copilot/tools/workspace_files.py:0-0
Timestamp: 2026-04-01T04:17:41.600Z
Learning: When reviewing AutoGPT Copilot tool implementations, accept that `readOnlyHint=True` (provided via `ToolAnnotations`) may be applied unconditionally to *all* tools—even tools that have side effects (e.g., `bash_exec`, `write_workspace_file`, or other write/save operations). Do **not** flag these tools for having `readOnlyHint=True`; this is intentional to enable fully-parallel dispatch by the Anthropic SDK/CLI and has been E2E validated. Only flag `readOnlyHint` issues if they conflict with the established `ToolAnnotations` behavior (e.g., missing/incorrect propagation relative to the intended annotation mechanism).
Applied to files:
autogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/config.py
📚 Learning: 2026-03-05T15:42:08.207Z
Learnt from: ntindle
Repo: Significant-Gravitas/AutoGPT PR: 12297
File: .claude/skills/backend-check/SKILL.md:14-16
Timestamp: 2026-03-05T15:42:08.207Z
Learning: In Python files under autogpt_platform/backend (recursively), rely on poetry run format to perform formatting (Black + isort) and linting (ruff). Do not run poetry run lint as a separate step after poetry run format, since format already includes linting checks.
Applied to files:
autogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/config.py
📚 Learning: 2026-03-16T16:35:40.236Z
Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12440
File: autogpt_platform/backend/backend/api/features/workflow_import.py:54-63
Timestamp: 2026-03-16T16:35:40.236Z
Learning: Avoid using the word 'competitor' in public-facing identifiers and text. Use neutral naming for API paths, model names, function names, and UI text. Examples: rename 'CompetitorFormat' to 'SourcePlatform', 'convert_competitor_workflow' to 'convert_workflow', '/competitor-workflow' to '/workflow'. Apply this guideline to files under autogpt_platform/backend and autogpt_platform/frontend.
Applied to files:
autogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/config.py
📚 Learning: 2026-03-31T15:37:38.626Z
Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12623
File: autogpt_platform/backend/backend/copilot/tools/agent_generator/fixer.py:37-47
Timestamp: 2026-03-31T15:37:38.626Z
Learning: When validating/constructing Anthropic API model IDs in Significant-Gravitas/AutoGPT, allow the hyphen-separated Claude Opus 4.6 model ID `claude-opus-4-6` (it corresponds to `LlmModel.CLAUDE_4_6_OPUS` in `autogpt_platform/backend/backend/blocks/llm.py`). Do NOT require the dot-separated form in Anthropic contexts. Only OpenRouter routing variants should use the dot separator (e.g., `anthropic/claude-opus-4.6`); `claude-opus-4-6` should be treated as correct when passed to Anthropic, and flagged only if it’s used in the OpenRouter path where the dot form is expected.
Applied to files:
autogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/config.py
📚 Learning: 2026-04-15T02:43:36.890Z
Learnt from: ntindle
Repo: Significant-Gravitas/AutoGPT PR: 12780
File: autogpt_platform/backend/backend/copilot/tools/workspace_files.py:0-0
Timestamp: 2026-04-15T02:43:36.890Z
Learning: When reviewing Python exception handlers, do not flag `isinstance(e, X)` checks as dead/unreachable if the caught exception `X` is a subclass of the exception type being handled. For example, if `X` (e.g., `VirusScanError`) inherits from `ValueError` (directly or via an intermediate class) and it can be raised within an `except ValueError:` block, then `isinstance(e, X)` inside that handler is reachable and should not be treated as dead code.
Applied to files:
autogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/config.py
📚 Learning: 2026-04-22T11:46:04.431Z
Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12881
File: autogpt_platform/backend/backend/copilot/config.py:0-0
Timestamp: 2026-04-22T11:46:04.431Z
Learning: Do not flag the Claude Sonnet 4.6 model ID as incorrect when it uses the project’s established hyphenated convention: `anthropic/claude-sonnet-4-6`. This hyphen form is the intentional, production convention and should be treated as valid (including in files like llm.py, blocks tests, reasoning.py, `_is_anthropic_model` tests, and config defaults). Note that OpenRouter also accepts the dot variant `anthropic/claude-sonnet-4.6`, so either form may be tolerated, but `anthropic/claude-sonnet-4-6` should be considered the standard to match project usage.
Applied to files:
autogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/config.py
📚 Learning: 2026-04-22T11:46:12.892Z
Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12881
File: autogpt_platform/backend/backend/copilot/baseline/service.py:322-332
Timestamp: 2026-04-22T11:46:12.892Z
Learning: In this codebase (Significant-Gravitas/AutoGPT), OpenRouter-routed Anthropic model IDs should use the hyphen-separated convention (e.g., `anthropic/claude-sonnet-4-6`, `anthropic/claude-opus-4-6`). Although OpenRouter may accept both hyphen and dot variants, treat the hyphen-separated form as the intended, correct codebase-wide convention and do not flag it as an error. Only flag the dot-separated variant (e.g., `anthropic/claude-sonnet-4.6`) as incorrect when reviewing/validating model ID strings for OpenRouter-routed Anthropic models.
Applied to files:
autogpt_platform/backend/backend/copilot/service.pyautogpt_platform/backend/backend/copilot/sdk/env_test.pyautogpt_platform/backend/backend/copilot/baseline/service.pyautogpt_platform/backend/backend/copilot/executor/processor.pyautogpt_platform/backend/backend/copilot/sdk/env.pyautogpt_platform/backend/backend/copilot/config_test.pyautogpt_platform/backend/backend/copilot/executor/processor_test.pyautogpt_platform/backend/backend/copilot/config.py
🔇 Additional comments (2)
autogpt_platform/backend/backend/copilot/executor/processor.py (1)
513-519: Nice wiring of capability-aware mode routing.Passing
thinking_available=config.thinking_availableintoresolve_use_sdk_for_modekeeps transport capability enforcement in a single decision point.autogpt_platform/backend/backend/copilot/executor/processor_test.py (1)
137-179: Strong coverage for the new kill-switch semantics and legacy default behavior.These tests cleanly pin the forced-baseline path (including warning visibility) and protect the default routing behavior when
thinking_availableis not passed.
… simulator Round-3 self-review follow-up. After the previous commit (2fb6255) gave both transport branches a non-None ``extra_body`` value, the ``if extra_body is not None: create_kwargs["extra_body"] = extra_body`` guard inside the retry loop became dead code and the ``dict[str, Any] | None`` annotation became wider than the actual value. Inline ``"extra_body": extra_body`` directly into ``create_kwargs`` and narrow the annotation.
…meout_s under local transport Cursor Bugbot caught: ``_LLM_TIMEOUT = 30`` was sized for OpenRouter latency. Under ``CHAT_USE_LOCAL=true`` on CPU-only hardware (the launch target for the no-API-key install), small JSON extractions on 0.6B–3B Ollama models routinely take 30–120+ seconds, so the asyncio.wait_for timeout fires before the model finishes and every Tally form submission raises ``TimeoutError``. Track ``chat_cfg.local_request_timeout_s`` (the same knob that gates the AsyncOpenAI HTTP-client timeout, default 1800s) under local transport so the application-layer wait_for and the underlying HTTP timeout stay aligned. OpenRouter/cloud paths keep the original 30s budget.
…_understanding under local transport Cursor Bugbot caught the last call site that wasn't forwarding the ``options.num_ctx`` hint under local transport. Without it, Tally's ``extract_business_understanding`` lets Ollama's OpenAI shim cap context at its 4 k default (ollama/ollama#2714), silently truncating long form submissions and producing garbage JSON. Mirror the pattern already in baseline/service.py + simulator.py + activity_status_generator.py: under ``CHAT_USE_LOCAL=true`` add ``extra_body={"options": {"num_ctx": chat_cfg.local_num_ctx}}`` to the chat-completions kwargs. Non-Ollama OpenAI-compat backends ignore unknown ``options`` keys, so the forward is safe across the local stack.
…nd-cache architecture test The factor-out helper introduced in 2fb6255 inherited the same process-wide-@cached(...) loop-binding caveat the architecture test in ``backend/util/architecture_test.py`` already grandfathers for the parent ``get_openai_client``. Add it to ``_KNOWN_OFFENDERS`` so CI stays green; the underlying migration to ``per_loop_cached`` will lift both together. CI failure was: ``test_no_process_cached_loop_bound_clients`` on Python 3.11/3.12/3.13.
…T per call Cursor flagged that ``simulator.py`` and ``activity_status_generator.py`` assigned the module-level ``_OPENROUTER_INCLUDE_USAGE_COST`` dict to ``extra_body`` by reference, while ``baseline/service.py`` already wraps it in ``dict(...)``. The OpenAI SDK treats ``extra_body`` as opaque pass-through, but an intermediate layer mutating the dict would leak across coroutines and corrupt every future call. Match baseline's defensive ``dict(_OPENROUTER_INCLUDE_USAGE_COST)`` idiom in both call sites so all three OpenRouter spots use the same shape.
…al transport Cursor flagged that ``generate_activity_status_for_execution`` silently discarded the caller's ``model_name`` argument under ``CHAT_USE_LOCAL=true`` — any explicit reroute (admin override, test fixture, future-feature consumer) was clobbered by ``chat_cfg.title_model``. Only auto-override when the caller left ``model_name`` at its cloud-routing default sentinel (``_DEFAULT_MODEL_NAME = "gpt-4o-mini"``); explicit caller values pass through untouched. Mirrors the pattern ``ChatConfig._apply_local_aux_models`` already uses for the ``title_model`` / ``simulation_model`` field defaults.
github-actions
Bot
added
the
conflicts
|
This pull request has conflicts with the base branch, please resolve those so we can evaluate the pull request. |
|
|
Bentlybro
left a comment
Bentlybro
left a comment
There was a problem hiding this comment.
I tested locally and it works so approving 😄
github-project-automation
Bot
moved this from 🆕 Needs initial review
to 👍🏼 Mergeable
in AutoGPT development kanban
Resolve 4 conflicts in copilot config + clients: - copilot/config.py: combine local-transport TransportProfile system (HEAD) with dev's _host_matches helper, ANTHROPIC_OPENAI_COMPAT_BASE_URL constant, and main/aux client credential properties. Update _DEFAULT_TITLE_MODEL default to anthropic/claude-haiku-4-5 (dev) so direct-Anthropic deployments pass aux validation. Merge SDK vendor validator to use transport-profile constraint plus dev's bare-slug validation. Add local-transport fast-path to _validate_aux_client_for_direct_main so aux under Ollama isn't held to the Anthropic-only title constraint. - copilot/service.py: keep dev's _get_main_client + _get_aux_client split with the back-compat _get_openai_client alias and reset_clients helper, while preserving HEAD's local-transport timeout extension on both clients. - copilot/baseline/service.py: combine HEAD's local-transport branch (num_ctx, no OR cost params) with dev's openrouter_active vs direct-Anthropic split for reasoning vs anthropic_thinking_extra_body. - copilot/config_test.py: keep both branches' new tests; update test_cloud_transport_does_not_inherit assertion to the new default title model. Boy-scout: drop unused `now = datetime.now()` in two store/db_test.py helpers and an unused CopilotPermissions import in chat/routes.py flagged by ruff F841. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
github-actions
Bot
removed
the
conflicts
|
Conflicts have been resolved! 🎉 A maintainer will review the pull request shortly. |
| extra_body.update(reasoning_param) | ||
| else: | ||
| extra_body = {} | ||
| # Direct mode (non-OR, non-local): use native Anthropic thinking param. |
There was a problem hiding this comment.
OpenRouter stream_options sent to local backends
Medium Severity
Under local transport with the default use_openrouter=True, openrouter_active returns True (because api_key is truthy, base_url starts with "http", and use_openrouter is True). The extra_body logic at line 676 correctly branches on config.transport.name == "local", but the subsequent config.openrouter_active checks at lines 709 and 740 are not guarded by a local-transport exclusion. This causes OpenRouter-specific stream_options to be sent to Ollama/vLLM and cost tracking to use the wrong mode (treating local responses as OpenRouter). Stricter OpenAI-compat backends may reject the unexpected parameter.
Additional Locations (1)
Reviewed by Cursor Bugbot for commit 3e4f669. Configure here.
…changed After the dev merge, ``normalize_model_for_transport`` was only special-casing the OpenRouter branch; everything else (subscription, direct_anthropic, local) fell into the Anthropic-only validator and raised ``ValueError`` for any non-``claude-*`` bare slug. Local backends (Ollama, vLLM, LM Studio, ...) use operator-chosen slugs like ``llama3.1:8b-instruct-q4_K_M`` that don't fit the ``vendor/model`` convention — and rewriting them to strip a missing prefix would 404 every call against the local endpoint. Adds ``local`` to the same passthrough branch as ``openrouter`` so local slugs flow through unchanged. Caught on a fresh Proxmox VM running the ``--with-ollama`` installer: the very first chat raised ``'local' transport requires an Anthropic model slug, got model='llama3.1:8b-instruct-q4_K_M'``. Regression test pins the local-passthrough behaviour against a bare ``llama3.1:8b-instruct-q4_K_M`` slug, sitting alongside the existing cloud-transport cases. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
A first-time user on CPU-only hardware will see AutoPilot appear hung for 10-15 minutes on the first turn while llama.cpp prefills the ~3 k-token system prompt. The existing troubleshooting entry only covered the 5-15 s model-load delay; this expands it to set expectations on the ongoing per-turn prefill cost and points at the Ollama log line that signals "prefill done, generating". Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 2 potential issues.
There are 3 total unresolved issues (including 1 from previous review).
❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.
Reviewed by Cursor Bugbot for commit 3eef535. Configure here.
| "CHAT_BASE_URL + CHAT_API_KEY (local). See " | ||
| "docs/platform/copilot-local-llm.md." | ||
| ) | ||
| model = chat_cfg.title_model |
There was a problem hiding this comment.
Tally extraction model silently changed for cloud transports
Medium Severity
extract_business_understanding now uses chat_cfg.title_model (defaulting to "anthropic/claude-haiku-4-5") for all transports, replacing the previous hardcoded "openai/gpt-4o-mini". The PR description states "Non-local transports keep their existing behaviour exactly", but this model change applies to the cloud (OpenRouter) path too. Claude Haiku is a different model with different pricing and output characteristics than GPT-4o-mini, so existing cloud deployments see a silent model swap for every Tally business-understanding extraction.
Reviewed by Cursor Bugbot for commit 3eef535. Configure here.
| # ``activity_status_generator.py``. Defends against intermediate | ||
| # layers ever mutating ``extra_body`` and corrupting the shared | ||
| # module-level dict for every future call. | ||
| extra_body = dict(_OPENROUTER_INCLUDE_USAGE_COST) |
There was a problem hiding this comment.
Simulator records wrong provider label under local transport
Low Severity
The new local-transport branches in _call_llm_for_simulation and generate_activity_status_for_execution route LLM calls through the local Ollama/vLLM client, but downstream cost-tracking still records provider="open_router" (existing hardcoded value in _track_simulator_cost). Previously these callers returned None and skipped when no OpenRouter key was present, so the hardcoded label was never reached under the equivalent scenario. Now it is, producing misleading cost attribution in PlatformCostLog for local-transport deployments.
Additional Locations (1)
Reviewed by Cursor Bugbot for commit 3eef535. Configure here.
3 participants
Why
backend/backend/copilot/config.py(ChatConfig) hard-wires the AutoPilot chat path to OpenRouter / Anthropic via OpenRouter-format model slugs. Operators trying to run AutoPilot without a Claude / OpenAI / OpenRouter key today hit two failure modes:CHAT_API_KEYis empty, so AutoPilot is unusable.CHAT_BASE_URLat a local OpenAI-compat endpoint, the SDK ("extended thinking") mode still fires for default requests, the CLI tries to speak Anthropic's wire protocol against Ollama, and they get an opaque 500 instead of a graceful downgrade.This is a launch-blocker for the "no API keys required" install story being validated on the multi-OS test bed.
What
Adds a fourth, explicit
localtransport toChatConfig.effective_transport(and aTransportProfiledescriptor that consolidates per-transport behaviour for all four). WhenCHAT_USE_LOCAL=true:CHAT_BASE_URLover OpenAI-compatible HTTP. Wins over OpenRouter / subscription (an inherited token in CI / dev shells doesn't silently re-route).transport.supports_sdk == False. Requests withmode='extended_thinking'are downgraded to fast at the request layer with a logged WARNING;build_sdk_envraisesRuntimeErrorif anything still tries to construct the SDK env (defensive).CHAT_*_MODELaccepts bare Ollama-style names (llama3.1:8b-instruct-q4_K_M) without ananthropic/*placeholder.OPENAI_API_KEYset for graphiti / embedders no longer silently binds to the local Ollama endpoint as the bearer token.title_model/simulation_modelleft at their cloud defaults inheritfast_standard_modelso operators only set one model slug instead of three.num_ctx—local_num_ctxfield (default 32 768) forwarded for OpenAI-compat backends that honoroptions.num_ctxin the request body. (Ollama itself doesn't — its OpenAI shim ignores it; setOLLAMA_CONTEXT_LENGTHon the systemd unit instead. The bundled installer's--with-ollamadoes this for you. Documented.)local_request_timeout_s(default 1 800 s) replaces the OpenAI client's 600 s default, so CPU-only local backends don't fail one turn before the model finishes.Non-
localtransports keep their existing behaviour exactly —use_localdefaults False, the newthinking_availablekwarg defaults True, every cloud path is byte-identical.How
backend/copilot/config.pyTransportProfiledataclass +_TRANSPORT_PROFILEStable for all four transports; newuse_local/local_num_ctx/local_request_timeout_sfields;transportproperty;effective_transportadds"local"; vendor validator + api_key fallback + aux-model derivation rewritten as model_validators that read the profile;thinking_availablekept as backwards-compat alias fortransport.supports_sdkbackend/copilot/baseline/service.pyextra_body.options.num_ctxfromconfig.local_num_ctxonly under the local transportbackend/copilot/service.py_get_openai_client()usesconfig.local_request_timeout_sunder the local transportbackend/copilot/executor/processor.pyresolve_use_sdk_for_modeacceptsthinking_available, downgrades extended_thinking → fast with WARNING log when the transport doesn't support the SDKbackend/copilot/sdk/env.pyRuntimeErrorwhen called under a no-SDK transportbackend/.env.defaultCHAT_USE_LOCAL/CHAT_BASE_URL/CHAT_API_KEYfor the local-LLM operatorinstaller/setup-autogpt.sh--with-ollama/--ollama-model=/--ollama-host=flags; bootstrap_ollama installs Ollama + setsOLLAMA_HOST=0.0.0.0:11434+OLLAMA_CONTEXT_LENGTH=32768via systemd drop-in; pulls the recommended default model (llama3.1:8b-instruct-q4_K_M); writesbackend/.envwith the rightCHAT_USE_LOCALblock.--ollama-host=URLskips the install for operators wiring an existing remote endpointdocs/platform/copilot-local-llm.mdnum_ctxgotcha, and a verification checklistdocs/platform/SUMMARY.mdTestTransportProfile(shape per transport),TestApiKeyFallback(local refuses OPENAI_API_KEY fallback; openrouter still falls back),TestLocalAuxModels(auto-derive + explicit override + cloud transports unaffected),TestLocalTransport(precedence over subscription/openrouter, validator skip, env var pickup),TestBuildSdkEnvLocalTransportGuard(defensive RuntimeError), processortest_thinking_unavailable_forces_baseline+test_thinking_available_default_preserves_legacy_behaviourValidation
End-to-end on a fresh Ubuntu 24.04 cloud VM:
setup-autogpt.sh --with-ollamafrom a clean repo → Ollama installed,OLLAMA_CONTEXT_LENGTH=32768set, model pulled,backend/.envwired, compose up.CHAT_USE_LOCAL=true,CHAT_BASE_URL=http://192.168.1.185:11434/v1,CHAT_FAST_STANDARD_MODEL=qwen3:0.6b,CHAT_TITLE_MODEL=qwen3:0.6b(auto-derived).python -c "from backend.copilot.config import ChatConfig"confirmedtransport.name == "local",transport.supports_sdk is False,local_num_ctx == 32768,local_request_timeout_s == 1800./copilot→ sent "What is 17 multiplied by 23? Reply with only the numeric answer." → AutoPilot responded391(correct), chat auto-titled "391" (title-model derivation working).Using baseline service (mode=default)(the SDK downgrade path); Ollama log: 200 from/v1/chat/completions.(Performance note: the test VM is 4-core CPU only — qwen3:0.6b under the AutoPilot tool-call loop took 37 minutes on it. On any GPU host, or with a smaller system prompt, this is interactive. Hardware constraint, not a code defect.)
Out of scope (separate PRs)
localhost:8006build-time URL bake-in that prevents external-IP UI access without an SSH tunnel — separate frontend PR.REDIS_ARGS=--requirepass ${GRAPHITI_FALKORDB_PASSWORD:-}boots with mis-quoted args when password is empty) — separate platform-infra PR. Tolerated here because graphiti is gated by an LD flag that defaults off in self-hosted.setup.agpt.co/install.shDocker bootstrap. The script today hard-fails when Docker is missing despite docs claiming "no prerequisite setup needed" — separate gap, separate PR.Checklist
feat(backend/copilot))docs/platform/copilot-local-llm.md) + linked from SUMMARYNote
Medium Risk
Moderate risk because it changes core chat routing/credential resolution and propagates new local-LLM behavior into multiple LLM-backed helpers (chat, simulator, onboarding extraction, embeddings), plus installer automation that edits system services/env files.
Overview
Enables running AutoPilot against a self-hosted OpenAI-compatible endpoint by adding a new
localtransport (CHAT_USE_LOCAL,CHAT_BASE_URL,CHAT_API_KEY) with centralized per-transport behavior viaTransportProfile(SDK availability, api-key fallback policy, and aux-model defaults).Updates chat execution to downgrade
extended_thinkingto baseline when the SDK can’t run (local), adds local-specific request tuning (options.num_ctx, longer OpenAI-client timeouts), and makes helper features (simulator, activity-status generation, Tally extraction, store embeddings) respect the local transport and avoid OpenRouter-only request fields.Adds operator-facing support:
.env.defaultdocumentation, a newdocs/platform/copilot-local-llm.mdguide linked fromSUMMARY.md, andsetup-autogpt.sh --with-ollamaautomation to install/probe Ollama, pull a model, and idempotently write the required backend.envblock.Reviewed by Cursor Bugbot for commit 3eef535. Bugbot is set up for automated code reviews on this repo. Configure here.