Skip to main content

AgentKit

A package manager for agent capabilities.

AgentKit turns messy setup into deterministic recipes so agents can install tools, models, and workflows without guessing commands.

npm i -g @claudioabierto/agentkit
~75% fewer setup tokens
1–2 commands to install
GitHub‑native registry

AgentKit is Homebrew for agent capabilities — but each “package” is a full workflow (install, configure, validate).

What is AgentKit?​

AgentKit is a capability package manager for AI coding agents. It sits between package managers (brew/npm/pip) and agent runtimes. Agents can search, inspect, and add capabilities—deterministic, multi‑step recipes hosted in a GitHub‑native registry.

Why it matters​

Less guessing

Skip the “try a command → error → retry” loop.

Deterministic installs

Recipes make setup repeatable, reviewable, and auditable.

Token savings

Fewer guess loops means fewer LLM tokens (and $$$)!

Auditability

Every step is explicit, reviewable, and reproducible.

How it works​

1) Search

Find the capability you need in the registry.

2) Inspect

Review every step before anything runs.

3) Add

Execute the deterministic plan with logs.

agentkit search web-automation
agentkit inspect web-automation
agentkit add web-automation

Use cases​

Web automation

  • install chromium
  • install puppeteer
  • safety check

Media transcription

  • install ffmpeg
  • install yt-dlp
  • download video
  • transcribe

PII redaction

  • install presidio
  • configure analyzers
  • run redaction

Example capabilities​

web-automationmedia-transcriptionpresidiotensorflow-cputensorflow-gpu

Repos​