Commit History
-
refactor(agent): enable immediate response instruction in supervisor prompt
ed2f57e -
refactor(api): switch LLM provider from MiniMax to Google Generative AI
9236a37 -
fix(config): update MiniMax model identifier to MiniMax-M2.5-highspeed
4ad0ce1 -
refactor(api): add LLM abstraction layer to replace hardcoded model strings
Replace hardcoded "google_genai:gemini-3-flash-preview" model references across all agent modules (supervisor, websearch, answer extractor, ocr reader) with configurable get_llm() function from agent.api.api modulebcdc55d -
feat(tools): add http_get tool for web content fetching
- Import http_get module in agent.py - Register http_get in supervisor agent tools list - Provide HTTP GET capability for retrieving remote resourcesc36efba -
feat(tools): add ocr_reader and list_files tools to expand agent capabilities
- Add ocr_reader tool to extract text from image and photo attachments - Add list_files tool for directory listing functionality - Update file_downloader with User-Agent headers to avoid request blocks - Extend supervisor prompt with image handling and insufficient information response7a0b5ad -
feat(agent): add file URL support to agent query handling
Add optional `file_url` parameter to `run()` and `BasicAgent.__call__()` methods. When a task_id is available, construct and pass the file URL to the agent alongside the question text, allowing the supervisor to access attached files during processing.01faebd -
refactor(agent): remove unused websearch_agent and update supervisor instructions
Remove the unused `websearch_agent` import and tool from supervisor agent configuration, keeping only `web_search_agents`. Update system prompt to instruct the agent to respond immediately once answer is found and avoid unnecessary searching to prevent exceeding action step limits. Also apply code formatting improvements to app.py.c82d598 -
feat(run): add retry mechanism with error handling to agent entry point
Implement retry logic in the run() function with max_retries parameter (default 3). On failure, append error context to help the agent retry with a simpler approach. Add recursion_limit config of 30 to prevent infinite loops. Also update answer extractor output to wrap answer in quotes for better visibility.97fc18c -
feat(agent): add answer extraction module and integrate into run function
Add new answer_extractor module that processes agent responses to extract clean answers. Refactor main entry point to use the simplified run() function instead of interactive loop.da8c40b -
chore(agent): remove verbose research complete banner from web search output
b70637a -
refactor(agent): split websearch agent into separate modules
Split monolithic websearch.py into websearchagent.py and websearchagents.py to improve code organization and maintainability. The supervisor agent now has access to both individual websearch_agent and aggregated web_search_agents.a394be7
-
refactor(agent): prioritize accuracy in math solver tool instructions
Clarify that accuracy is the primary concern when using the math_solver tool for checking calculations. The instruction now explicitly states accuracy as the most important factor, ensuring the supervisor agent emphasizes precision over speed or convenience when routing math-related queries.0019780 -
chore(agent): enhance supervisor prompts and increase websearch recursion limit
Add "any" and "must" to make supervisor routing directives more emphatic and increase websearch recursion_limit from 3 to 6 to allow more thorough search results314810a -
fix(websearch): add recursion limit and error handling for search agent
Add recursion_limit=3 configuration and GraphRecursionError handling to prevent infinite loops in the web search agent. The agent now returns a partial result message when the recursion limit is reached instead of crashing.cee18af -
chore: remove __pycache__ bytecode files from version control
2f08174 -
feat(agent): implement supervisor agent with tool routing and math solver
- Replace single websearch agent with supervisor agent architecture - Add math_solver tool using sympy for calculation queries - Implement tool routing logic (math vs web search) in supervisor prompt - Add interactive chat loop with conversation history - Enhance terminal output with colored logging (colorama) - Update dependencies: sympy, colorama40deb66 -
feat(agent): add web search agent with LangGraph and Tavily integration
- Created agent module with websearch_agent and run() entrypoint - Implemented Tavily search tool (max_results=10) - Updated Gradio app to use the new agent instead of fixed answers - Added dependencies: langchain, langchain-google-genai, langgraph, tavily-python - Added .env.example for API key configuration (GOOGLE_API_KEY, TAVILY_API_KEY) - Updated README with setup and usage documentation - Removed old template files (main.py, pyproject.toml)a877f54