minion-ai 0.1.3

A simple agentic framework with observability and evals baked in

A lightweight, provider-agnostic agentic framework.
Build AI agents that think, use tools, and delegate to sub-agents — with observability baked in.

---

Install

pip install minion-ai

Quick Start

import minions
from minions.demo_tools import read_file, list_files

def search_web(query: str) -> str:
    """Search the web and return results.

    Args:
        query: The search query.
    """
    # your implementation
    ...

minions.init(api_key="your-api-key")  # or set ANTHROPIC_API_KEY / OPENAI_API_KEY env var

agent = minions.Minion(
    model="anthropic/claude-opus-4",
    tools=[search_web, read_file],
)

result = agent("Summarise the contents of report.pdf")
print(result)

Supported Providers

Any provider supported by LiteLLM works out of the box — just change the model string:

minions.Minion(model="openai/gpt-4o", ...)
minions.Minion(model="anthropic/claude-opus-4", ...)
minions.Minion(model="gemini/gemini-2.5-pro", ...)

API keys are read from the standard env vars (OPENAI_API_KEY, ANTHROPIC_API_KEY, etc.) or passed via minions.init().

Sub-Agents

Minion can spawn sub-agents to parallelise large tasks:

agent = minions.Minion(
    model="anthropic/claude-opus-4",
    tools=[read_file, list_files],
    allow_sub_agents=True,
    secondary_model="anthropic/claude-haiku-4-5",  # cheaper model for sub-tasks
)

result = agent("Summarise every file in the /reports directory")

The agent automatically delegates when it sees 3+ independent items to process.

Tracing & UI

Enable tracing to record every run, turn, and tool call to a local SQLite database, then inspect them in the built-in dashboard.

pip install minion-ai

minions.init(
    api_key="...",
    tracing=True,
    project="my-project",   # required when tracing=True
)

agent = minions.Minion(model="openai/gpt-4o", tools=[...])
agent("Do something interesting")

Launch the dashboard:

minion ui

The UI groups traces by project, lets you drill into every turn and tool call, and shows token usage and cost estimates in real time.

Remote tracing (team server)

To send traces to a shared minion-ui server instead of a local file, add trace_url and a project-scoped token (create one in the dashboard under a project's Settings → API Tokens):

minions.init(
    tracing=True,
    project="my-project",
    trace_url="https://traces.mycompany.com",
    tracing_secret_token="mni_xK9mP2...",
)

In this mode nothing is written locally — runs are pushed over HTTP to the server. Tracing never raises: if the server is unreachable the push is skipped and your agent keeps running.

Running the server

The dashboard server ships as a container:

docker compose up            # SQLite (default), data in a named volume

For a team-scale deployment, point it at Postgres via DATABASE_URL (SQLite stays the default when it's unset):

docker compose -f docker-compose.postgres.yml up          # self-hosted Postgres
docker compose -f docker-compose.managed-postgres.yml up  # RDS / Supabase / Neon

See docs/ for full hosting, remote-tracing, and image-publishing guides.

Building Tools

Any Python function with a docstring and typed args works as a tool:

def get_weather(city: str, unit: str = "celsius") -> str:
    """Get the current weather for a city.

    Args:
        city: Name of the city.
        unit: Temperature unit, either 'celsius' or 'fahrenheit'.
    """
    ...

agent = minions.Minion(model="openai/gpt-4o", tools=[get_weather])

See minions/demo_tools.py for more examples.