Quick start¶

Get your first agent team running in 5 minutes. No cloud account. No API key required.

Step 1 — Install¶

pip install antcrew

Step 2 — Run your first pipeline¶

Option A — Simulated LLM (instant, zero cost, no setup):

antcrew run --model simulated "Build a REST API for user authentication"

This runs the full pipeline — BA, PM, backend dev, QA, reviewer — using a deterministic simulated model that produces the same artifacts every time. The content is fake (no real AI), but the artifact structure is real — useful for testing pipelines and CI, not for actual development output.

No LLM is bundled with antcrew. The simulated model is a built-in test stub. For real AI output, use Option B (local Ollama) or Option C (cloud API key). If you want real AI with zero setup, use antcrew-platform's managed tier — it provides the LLM.

Option B — Fully local with Ollama (no API key, no data leaves your machine):

# 1. Install Ollama from https://ollama.com, then:
ollama pull llama3

# 2. Run antcrew against it
antcrew run --model ollama:llama3 "Build a REST API for user authentication"

Option C — Cloud model:

export ANTHROPIC_API_KEY=sk-ant-...
antcrew run --model claude "Build a REST API for user authentication"

Step 3 — See what was produced¶

# View the run summary
antcrew inspect <run-id>

# Shows per-agent: prompt, response, tokens, cost, governance hash

The run ID is printed at the end of every antcrew run. The inspect output includes every agent's input and output — a full trace of the decision chain.

Step 4 — From Python¶

from antcrew import DevTeam
from antcrew.models import OllamaModel, AnthropicModel, SimulatedLLM

# Local — no API key
model = OllamaModel("llama3")
# Or: model = SimulatedLLM()
# Or: model = AnthropicModel("claude-sonnet-4-6")

team = DevTeam(model=model)
result = team.run("Build a REST API for user authentication")

# Typed artifacts — not dicts
prd = result.state["prd"]
print(prd.title)
print(prd.tech_stack)

tickets = result.state["tickets"]
for ticket in tickets:
    print(f"  [{ticket.priority}] {ticket.title}")

print(f"\nTotal cost: ${result.cost_usd:.4f}")   # 0.0 with Ollama

Step 5 — Replay the trace¶

# Replay every agent call to detect model drift
antcrew trace replay <run-id>

Replay reruns all agent calls and compares outputs to the original. If a model update changed behavior, replay will tell you which agents diverged.

What's next¶

Try other teams — FullStackTeam, ResearchTeam, ContentTeam, LegalReviewTeam
Typed artifacts reference — Contracts & artifact types
Add a human checkpoint — HITL local approvals
Use 100+ LLM providers — Providers
Run in CI — EvalSuite regression testing
Add semantic memory — pip install "antcrew[memory]" then pass ChromaMemory() to any team
Governance hash — cite agent configs in papers or pin in audit pipelines: antcrew inspect <id>

Optional cloud layer¶

The antcrew SDK runs entirely locally — no cloud account needed. When your team needs shared runs, remote HITL, and a dashboard, deploy antcrew-platform.