crawfish 0.3.0

Agents for bulk work over your data: Source -> Batch -> Aggregator -> Router -> Sink, run locally.

Crawfish

Agents for bulk work over your data. Author a pipeline as a directory — Source → Batch (fan-out) → Aggregator (reduce) → Router (branch) → Sink — and run it locally via claude -p with zero API key. Typed, versioned, benchmarked.

Think dbt / Airflow for agents, not another chatbot SDK.

Install

pip install crawfish

Prefer an isolated CLI? Use uv tool install crawfish or pipx install crawfish.

Quickstart

craw init my-app && cd my-app
craw dev definitions/triage-bot -i project=acme -i "ticket_body=login is broken"

craw init scaffolds a working triage-bot example. craw dev runs it on a mock runtime — no key, no cost. The Source fans the item out, a Definition team runs per item, an Aggregator reduces, a Router branches, and a Sink writes — and the Output comes back typed.

Documentation

📖 neal-kotval.github.io/crawfish

• Getting started
• Tutorial — build the triage bot end to end
• Concepts — the directory model, runtimes, the security boundary
• API reference

Links

• Source: github.com/Neal-Kotval/crawfish
• Issues: github.com/Neal-Kotval/crawfish/issues
• Changelog: CHANGELOG.md

License

Apache-2.0.