Fim One
Open-source agent platform for Global × China enterprises — wire every system through one agent core. Self-hosted, any LLM.
Last verified:
What is Fim One?
FIM One is an AI-powered Python framework for building agents that dynamically plan and execute complex tasks across enterprise systems. It is a provider-agnostic, source-available platform that bridges your existing systems proactively—reading databases, calling APIs, and pushing notifications through a unified AI interface. Unlike rigid workflow builders, FIM One uses dynamic planning to orchestrate across ERP, CRM, OA, HR, finance systems, and IM platforms without requiring you to replicate logic.
The platform offers three delivery modes: Standalone (a general-purpose AI assistant for chat, code execution, and knowledge base Q&A), Copilot (AI embedded in your existing UI via iframe/widget/embed), and Hub (central cross-system orchestration where all systems connect through one agent core). The Hub architecture is the key differentiator, serving as a central portal where connectors act as standardized bridges to any system whether it's SAP, PostgreSQL, or custom databases.
FIM One is designed for Global × China enterprises that need to wire every system they already run—including global SaaS and the China stack like Feishu, WeCom, DingTalk, DM, and Kingbase—through one agent core. It supports any OpenAI-compatible LLM provider (OpenAI, DeepSeek, Claude, Qwen, Gemini) and local/self-hosted models (Ollama, vLLM, LocalAI, LM Studio). The AI Builder lets you create Connectors and Agents using natural language, and it supports 6 languages with automated translation.
Fim One pricing
Pricing model: Freemium
FIM One offers a Cloud version at cloud.fim.ai with zero setup (no Docker, no API keys, no configuration) for early access. The self-hosted version is source-available under a license permitting internal organizational use, modification, custom development, and distribution with the license intact. Commercial licensing is available for inquiries via GitHub issue. No specific pricing tiers or dollar amounts are publicly listed on the website.
Fim One pros
- Provider-agnostic—works with any OpenAI-compatible LLM
- Supports both cloud APIs and self-hosted local models
- Three delivery modes: Standalone, Copilot, and Hub
- AI Builder creates connectors using natural language
- Import OpenAPI specs to auto-generate connectors
- MCP protocol support for Model Context Protocol servers
- Dynamic DAG planning with parallel execution
- ReAct reasoning engine for focused queries
- Auto mode routes queries to optimal engine automatically
- Supports 10+ database types including Chinese databases
- Source-available license with broad organizational freedoms
- Zero Docker/API key setup available via Cloud version
- 6-language support with automated translation pipeline
- Apple Silicon (M1/M2/M3/M4) and ARM compatible
- FastAPI backend with Next.js/React TypeScript frontend
- Up to 3 rounds of automatic re-planning in DAG mode
- Read-only query execution with schema introspection
- Telegram/Lark/WeCom notification integration support
Fim One cons
- Not OSI-approved open source (source-available license)
- Multi-tenant SaaS offerings are restricted
- Competing agent platforms are prohibited
- White-labeling or removing branding is not allowed
- Requires Python 3.11+ and Node.js 18+ for local dev
- Minimum 2 GB RAM required (4 GB+ recommended)
- PostgreSQL required for multi-worker production deployments
- Self-hosting requires Docker 20+ and Compose v2
Frequently asked questions about Fim One
What LLM providers are supported?
FIM One works with any OpenAI-compatible LLM provider including commercial APIs like OpenAI, DeepSeek, Anthropic (Claude), Alibaba Qwen, and Google Gemini. It also supports local/self-hosted runtimes like Ollama, vLLM, LocalAI, and LM Studio that serve the OpenAI-compatible API format. Configuration uses three environment variables: LLM_API_KEY, LLM_BASE_URL, and LLM_MODEL.
Can I use local or self-hosted models?
Yes. FIM One connects to any endpoint implementing the OpenAI-compatible API format. Popular options include Ollama at localhost:11434/v1 with qwen2.5:14b, vLLM at localhost:8000/v1 with Qwen2.5-72B-Instruct, LocalAI at localhost:8080/v1 with llama3, and LM Studio. Set LLM_API_KEY to any non-empty string when the provider doesn't require authentication. All agent features work identically regardless of model location.
Is FIM One open source?
FIM One is released under a Source Available License, not OSI-approved open source. Permitted uses include internal organizational use, modification, custom development, distribution with intact license, and embedding in non-competing applications. Restricted uses include multi-tenant SaaS offerings, competing agent platforms, and white-labeling or removing branding. Commercial licensing inquiries should open a GitHub issue.
What are the system requirements?
Minimum requirements are Python 3.11+, 2 GB RAM, 1 GB free disk space, and Node.js 18+ for local development. For production, recommended specs are 4 GB+ RAM (especially for local embedding models), 2+ CPU cores, and PostgreSQL for multi-worker deployments. Docker alternative requires Docker 20+ and Compose v2 with no local Python or Node.js needed.
What is the difference between ReAct and DAG mode?
ReAct (Standard) uses a single reasoning loop: Reason, Act, Observe, repeat—best for focused queries, single-system lookups, and conversational tasks with sequential execution. DAG (Planner) has the LLM decompose goals into a dependency graph with independent steps running in parallel via asyncio—best for multi-step tasks, cross-system orchestration, and parallel data gathering with up to 3 rounds of automatic re-planning. Auto mode uses a fast LLM classifier to route queries optimally.
How do I add a new connector?
FIM One provides three code-free methods: 1) Import an OpenAPI spec (YAML, JSON, or URL) to auto-generate connectors with all actions, 2) AI chat builder to describe the API in natural language with the AI generating and iterating on connector configuration using 10 specialized builder tools, and 3) MCP protocol to connect any Model Context Protocol server directly. Database connectors auto-generate list_tables, describe_table, and query tools.
Can I embed FIM One into my existing system?
Yes. Copilot mode is designed for embedding via iframe (embed chat interface directly into any web page), widget (lightweight chat overlay on existing UI), or API (use FastAPI backend for fully custom integrations). In Copilot mode, AI works alongside users in their familiar interface for querying data, generating reports, and orchestrating actions without application switching.
What databases are supported?
For internal data: SQLite is the zero-config default for development, PostgreSQL is recommended for production and required for multi-worker setups. For connector targets: PostgreSQL, MySQL, Oracle, SQL Server, DM (Dameng), KingbaseES, GBase, and Highgo are all supported. Each database connector auto-generates three tools with schema introspection, AI-powered annotation, and read-only query execution included by default.
How does the translation system work?
FIM One supports 6 languages (English, Chinese, Japanese, Korean, German, French) with automated translation. Only edit English source files in frontend/messages/en/.json, docs/.mdx, and README.md. A pre-commit hook detects changes and translates via the project's Fast LLM incrementally. Translated files in messages/zh/, messages/ja/, docs/zh/, docs/ja/ are auto-generated and overwritten. Set up hooks with bash scripts/setup-hooks.sh or force full retranslation with uv run scripts/translate.py --all.
How can I contribute to FIM One?
FIM One welcomes code, documentation, translations, bug reports, and feature ideas. Read the Contributing Guide for setup and conventions, browse Good First Issues for newcomer tasks, and check Open Issues for bugs/feature requests. The Pioneer Program recognizes the first 100 contributors with a PR merged as Founding Contributors with permanent credits, a profile badge, and priority issue support. Get help via Documentation, Discord, GitHub Issues, or follow @FIM_One on Twitter/X.