Summary

FIM One provides an AI-powered agent runtime and hub to bridge disparate enterprise systems like ERP, CRM, and databases without infrastructure modification. It enables cross-system orchestration and automation, functioning as an embedded Copilot or a central integration Hub.

How It Works

The core approach utilizes LLM-driven dynamic DAG planning and concurrent execution. Goals are decomposed into runtime dependency graphs, allowing independent steps to run in parallel via asyncio with auto re-planning. Its ReAct agent features automatic error recovery and production-grade execution. Key innovations include progressive disclosure meta-tools reducing token usage by over 80% and a standardized connector architecture that auto-registers system actions as agent tools.

Quick Start & Requirements

• Primary install: Docker Compose (git clone ... && cd fim-one && cp example.env .env && docker compose up --build -d).
• Prerequisites: Python 3.11+, uv, Node.js 18+, pnpm, and an OpenAI-compatible LLM API key.
• Setup: Docker setup is straightforward; local development requires managing Python and Node.js environments.
• Links: Docs, Changelog, Roadmap, FAQ.

Highlighted Details

• Connector Hub: Integrates diverse systems (APIs, databases, MCP servers) via standardized connectors, supporting major SQL and Chinese legacy DBs.
• Dynamic DAG Planning & Concurrent Execution: LLM decomposes goals into runtime DAGs for parallel execution with auto re-planning.
• Intelligent Document Processing: Handles PDFs, DOCX, PPTX with vision-aware processing and image extraction.
• Full RAG Pipeline: Implements Jina embedding, LanceDB, hybrid retrieval, reranking, and inline citations.
• Multi-tenant Platform: Features JWT auth, org isolation, admin panel, and a marketplace for agents/skills.

Maintenance & Community

The project maintains active CI and encourages contributions via a "Pioneer Program." Community support is available via Discord and updates on X.

Licensing & Compatibility

Licensed under a non-OSI-approved "Source Available License." Permitted uses include internal deployment and distribution with license intact. Restricted uses cover multi-tenant SaaS, competing platforms, and white-labeling. Commercial inquiries should be directed via GitHub issues.

Limitations & Caveats

The "Source Available License" restricts SaaS and competing platform usage, potentially limiting broad commercial adoption. Setup requires specific Python and Node.js versions.