Maicel
A security-first agent operating system that grows with you.
What is Maicel?
Maicel is your personal AI assistant that runs entirely on your machine, with your data, using your API keys. It is not a SaaS product. There is no cloud service between you and your agents.
At its core, Maicel is an agent operating system. It manages agents the way an OS manages processes — with scheduling, resource control, security boundaries, and inter-process communication. Agents can be LLM-powered conversationalists, deterministic data processors, or anything in between.
Maicel starts small — a single chat agent that knows you — and grows organically. As you use it, you can create new agents, connect external tools, build reusable workflows, and accumulate a knowledge base. The system learns your preferences and adapts, but never without your awareness.
Core Principles
The Maicel Constitution — non-negotiable rules that every component follows.
Audit Everything
Every state-changing operation is logged. If it changes data, it gets audited. No exceptions, no silent mutations.
Security Fail-Closed
When a security check fails or encounters an error, the action is denied. Never allow on uncertainty.
Credentials Never Visible
No credential appears in logs, LLM prompts, error messages, or tracebacks. Encrypted at rest, decrypted only at the point of use.
MCP for External Services
All external integrations use the Model Context Protocol. Standard protocol, open ecosystem, no vendor lock-in.
Cost-Optimized
Always use the cheapest execution method first. Deterministic logic before Haiku, Haiku before Sonnet, Sonnet before Opus.
Open Source
MIT licensed, community-driven. You can read, audit, and modify every line of code that runs on your machine.
Why Open Source?
Trust
You can verify exactly what the code does. Every security claim is auditable. No black boxes, no hidden telemetry.
Privacy
Your data stays on your machine. Conversations, credentials, and knowledge never leave your control. The only external calls are to the LLM providers you choose.
Extensibility
Build your own agents, connectors, and workflows. Every component follows a Protocol interface that you can implement and swap.
Community
Contributions are welcome. Whether it is a new MCP connector, a bug fix, or a documentation improvement — the project grows through collaboration.
Technology
Maicel is built on proven, stable foundations with minimal dependencies.
Tech stack: Python 3.12+, SQLite (WAL mode), LiteLLM, MCP Protocol, AES-256-GCM, NixOS-style Config
- SQLite with WAL mode — fast, single-file database with concurrent read access. No database server to manage.
- LiteLLM — unified interface to multiple LLM providers. Switch between Anthropic, OpenAI, Google Gemini, and Ollama (local models) without code changes.
- NixOS-style config generations — every configuration change creates an immutable snapshot. Roll back to any previous state with a single command.
- MCP (Model Context Protocol) — the emerging standard for connecting AI agents to external tools. Built for interoperability.
Supported LLM Providers
| Provider | Models |
|---|---|
| Anthropic | Claude Opus, Sonnet, Haiku |
| OpenAI | GPT-4o, GPT-4o-mini, o1, o3 |
| Gemini 2.5 Pro, Flash | |
| Ollama (Local) | Llama, Mistral, Qwen, and more |
Getting Involved
Maicel is built in the open. Whether you want to use it, extend it, or contribute back — you are welcome.
- GitHub Repository — Source code, issues, and releases
- MIT License — Free to use, modify, and distribute
- Contributing — Pull requests welcome — all skill levels
- Community — Coming soon — Discord and forum
Like mycelium connecting a forest, Maicel connects your tools, knowledge, and workflows into a living network that grows with you.