Tool-calling AI agents that run locally.

Build once, run anywhere.

Your models, your tools, your machine.

from onsetlab import Agent

agent = Agent(
    model="qwen3-1.7b",
    tools=["Calculator", "DateTime"]
)

agent.connect_mcp("github", token="...")
result = agent.run("Summarize my PRs")

$ pip install onsetlab

What makes it different

REWOO + ReAct

Plans all steps upfront, executes in sequence, and falls back to step-by-step reasoning when a tool call fails.

MCP Servers

GitHub, Slack, Notion, filesystem, web search - connect any MCP server with a single line of config.

Self-Correcting

Wrong parameters? The agent retries with corrected inputs. Bad plan? ReAct fallback kicks in automatically.

Any Ollama Model

Qwen, Mistral, Hermes, Gemma - use whatever fits your hardware. No API keys, no cloud dependency.

See it in action

Watch how OnsetLab handles real queries from start to finish.

From install to deploy

Four simple steps to ship your agent.

1. Install
   pip install onsetlab

2. Configure
   Pick a model, add tools & MCP servers

3. Run
   agent.run("your task")

4. Deploy
   Export as Docker, YAML, or script

Ready to build?