AgentTrace Adds Observability and Self-Healing to AI Agent Pipelines

Summary

AgentTrace is an end-to-end observability SDK and dashboard for autonomous AI agent pipelines. It tracks multi-step tool calls and visualizes latency bottlenecks across a run. Its self-healing layer detects malformed LLM-generated tool arguments before execution, including incorrect value types, invented field names, and missing required schema fields. When validation fails, the project uses fast inference through Groq to repair the payload instead of allowing the workflow to crash. The Python decorator SDK uses Pydantic to validate tool inputs, while a FastAPI collector stores traces in SQLite. A Next.js and Tailwind CSS dashboard includes a payload-diff inspector for examining the original and repaired arguments. The repository also links to a live dashboard and FastAPI Swagger documentation, and provides a local startup command using Uvicorn.