Skip to content

antcrew

Multi-agent framework for Python. Typed outputs. Full trace. Works offline.

One command to your first agent team — no cloud account, no API key required:

pip install antcrew
antcrew run "Build a FastAPI auth service" --model ollama:llama3

Or three lines of Python:

from antcrew import QuickStart

result = QuickStart.dev().run("Build a FastAPI auth service")
print(result.state["prd"].title)       # typed PRD artifact
print(result.state["code_artifacts"])  # typed code files

What antcrew gives you

Capability Detail
Typed output contracts Every agent produces a Pydantic artifact — PRD, CodeArtifact, SecurityReport — not a raw dict. Downstream agents receive typed inputs; mismatches fail at definition time, not at runtime.
Trace & replay Every run writes a local TraceLog (SQLite). Replay any past run step-by-step, diff two runs, or re-run from a checkpoint.
Works 100% offline Ollama is a first-class provider. --model ollama:llama3 routes all LLM calls to your local instance — no network, no API key, no cost.
3 lines to first agent team QuickStart.dev().run("Build a FastAPI auth service") is the entire program.
Governance hash per agent Each agent turn is SHA-256 hashed (model config + inputs + outputs). The hash is stored in the TraceLog and exposed via antcrew inspect.
29 CLI commands antcrew run, antcrew inspect, antcrew trace replay, antcrew test, and 25 more — all local, no platform account needed.

How the pieces fit together

flowchart LR
    subgraph local["Your machine"]
        AC["antcrew SDK\npip install antcrew"]
        OL["Ollama / LM Studio\nlocal models"]
    end

    subgraph cloud_opt["Optional — cloud"]
        AP["antcrew-platform\nRuns · HITL · Dashboard"]
        PX["keybridge\nBYOK key gateway"]
        CL["Cloud LLMs\nAnthropic · OpenAI · Groq"]
    end

    DEV["👤 You"] -->|"antcrew run"| AC
    AC --> OL
    AC -.->|"if you deploy platform"| AP
    AP -.->|"with proxy mode"| PX
    PX -.-> CL
    AC -.-> CL

antcrew runs entirely on your machine. The platform is an optional cloud layer — your pipelines work without it.

Start here

pip install antcrew

# Simulated LLM — zero cost, deterministic, runs anywhere
antcrew run --model simulated "Build a user auth module"

# Fully local with Ollama (ollama pull llama3 first)
antcrew run --model ollama:llama3 "Build a user auth module"

→ 5-minute quick start

pip install antcrew
export ANTHROPIC_API_KEY=sk-ant-...
antcrew run --model claude "Build a user auth module"

→ LLM providers

from antcrew import DevTeam
from antcrew.models import OllamaModel

team = DevTeam(model=OllamaModel("llama3"))
result = team.run("Build a user auth module")
print(result.state["prd"].title)
print(result.cost_usd)   # 0.0 with Ollama

→ A full team in action

Components

antcrew (this package) is what almost everyone needs:

pip install antcrew

It ships with antcrew_engine built in — the autonomous EngineLoop that powers named-role teams. You don't need to install anything extra.

antcrew-platform is an optional cloud backend for teams that need multi-workspace runs, remote HITL, and a shared dashboard. → Platform docs

keybridge is a companion to antcrew-platform for teams that want their LLM API keys to stay on their own infrastructure. → Proxy docs