[UPDATE][locius][0.2.11]Vision-only clicking for chat widgets, shadow DOM and iframe support - #3708
Conversation
… DOM and iframe support
|
tags [agent assistant] are not configured in Admin V2. This PR can still be merged. Please add the missing tags in Admin V2 so they become available for taxonomy filtering. Currently supported tags: [audio automation browser chatbot coding collaboration crm customer-server data database decision-model design doc-process downloader ecommerce embedding engine erp esignature file-management finance finetune framework gaming gateway health home-theater human-resources image-editing indexer knowledge-base lifestyle live-streaming llm low-code marketing media-management memory messaging middleware monitoring network note-taking ocr personal-assistant photos project-management rag reading rerank security smart-home social-network speaker-diarization speech-to-text storage surveillance system-one testing text-to-audio text-to-speech translation travel utilities virtual-machine voice-cloning web-search website-builder workbench] |
|
Check passed, please wait for auto-merge. |
Update: Locius 0.2.10 → 0.2.11
Private AI agent that runs on the user's Olares. Source (MIT): https://github.com/Drlucaslu/locius
What's new
browser_locatefinds something on screen with vision alone — a lettered grid, then a zoomed numbered grid, then a red marker the model must confirm — andbrowser_click_atclicks there, optionally typing text and pressing Enter. Chat bubbles and chat windows inside iframes or widgets (Tidio, Intercom, Zendesk…) now work. Uses the local Olares model by default.browser_findandbrowser_looknow also see inside open shadow roots and iframes.Checks
python3 build.py locius;olares-cli chart lintpassespersona, same chart); tested live with the Tidio website chat bot and Amazon.sg customer service chatownersandvalues.yamlunchanged