V1-OpenAgent

OpenAgent v1.0.0

Open reference pipeline for agents that think before they act — by OpenGraph.tech. Public source: github.com/OpenGraph-AI/OpenAgent.

Highlights

• Five-stage pipeline with typed contracts between stages: Intent → Ambiguity → Clarifier → Planner → Executor. Each stage has a clear Pydantic-shaped boundary so you can test, mock, and debug in isolation.
• Streaming end-to-end so the UI can show phase-by-phase progress (intent, flags, clarifications, plan, execution).
• Clarifier that prefers the web over the user: optional web search (Exa) to auto-resolve ambiguity before asking; caps surfaced questions so the flow stays lightweight.
• Graceful degradation: missing optional services (Exa, Redis, RAG/PageIndex, etc.) skip those paths instead of failing hard — see .env.example for the full matrix.

Stack

• Backend: Python 3.10+, FastAPI (REST + WebSocket), Pydantic v2 schemas.
• Frontend: Static chat UI under public/ (and frontend/ assets where applicable).
• Deploy / run: python run.py with LLM_API_KEY minimum; optional Docker Compose and Vercel config for hosting patterns you already document in the README.

Docs & Assets

• README with quickstart, stage-by-stage "cookbook," and links to the 90-second walkthrough.
• Diagrams: public/stages.svg, public/mental-model.svg, branding in public/logo.svg.

For Operators

1. Clone the repo, pip install -r requirements.txt, copy .env.example → .env, set at least LLM_API_KEY (and optionally LLM_BASE_URL / LLM_MODEL).
2. Open the app at the static URL documented in the README (default dev server on port 8000).

Notes

• This release represents the initial public codebase as published on main (including merge with the repo's bootstrap .gitignore history). Treat semver as 1.0.0 for "first labeled release"; bump patch/minor as you ship fixes and features.