Skip to content

Frequently Asked Questions

Everything you need to know about MiMinions.

What is MiMinions?

MiMinions is an open-source Python framework for building autonomous AI agents. It enables developers to create, deploy, and manage agentic AI systems that can think, plan, and execute tasks. Built on top of pydantic_ai, it provides the building blocks for agentic systems: an LLM-powered Agent, a Tools registry, vector and markdown Memory, Workspaces, and MCP server integration.

How do I get started?

Getting started is easy. Install the framework with pip install miminions, then create your first agent in just a few lines of code. See the Getting Started guide for step-by-step instructions on building your first autonomous AI agent.

Is MiMinions free to use?

Yes — MiMinions is open-source and free to use. You can use it for personal projects, commercial applications, and contribute to its development. Check out the GitHub repository to see the source and contribute to the project.

Which LLM providers are supported?

MiMinions selects models through a ModelFactory. The default provider is OpenRouter (free openai/gpt-oss-20b:free model), and you can switch to OpenAI, Anthropic, Gemini, or an offline test model by passing provider= to create_minion:

from miminions.agent import create_minion

agent = create_minion("assistant", provider="anthropic")

Each real provider needs its API key in the environment (OPENROUTER_API_KEY, OPENAI_API_KEY, etc.). The test provider runs fully offline. See Agent for the full provider matrix.

Do I need a GPU?

No. MiMinions runs on CPU. The optional SQLite vector memory (pip install miminions[sqlite]) uses fastembed, which runs embeddings through ONNX on the CPU — no GPU or CUDA setup required. The LLM itself runs remotely via your chosen provider's API.

What safety features are included?

MiMinions takes a pragmatic, local-first approach to safety:

  • Sign-in gating on the CLI. Commands are wrapped with require_auth, so they refuse to run until you sign in with miminions auth signin. An opt-in public-access mode can relax this for trusted, local-only use.
  • Local-first data. Agents, workspaces, and memory persist under ~/.miminions/ on your own machine — nothing is sent to a third party beyond your chosen LLM provider.
  • Auditable data operations. The Data Management layer keeps an append-only transaction log of every write, giving you a complete audit trail of changes.
  • Per-channel allow-lists. The optional Gateway Runtime enforces an allow_from allow-list per channel: an empty list denies all senders, and you explicitly opt callers in (or use "*" to allow everyone).
Where can I find the documentation?

Full guides and API references live in the Documentation section, covering the Agent, Memory, Context Builder, Tools, Workspaces, Tasks & Workflows, Data Management, Gateway Runtime, and CLI & Chat modules.

Still have a question? Get in touch.