airegistry-cli 2.3.1

Install AI capabilities, not libraries.

AIRegistry is an open-source registry and dependency-free CLI for reusable, versioned AI building blocks: retrieval, memory, agents, prompts, tools, citations and validation.

Live registry · Issues · Contributing · Changelog

What is AIRegistry?

AI projects keep rebuilding the same small pieces: prompt builders, chunkers, retrievers, memory, validators, tool routers. AIRegistry packages them as small, readable, versioned source components that you copy into your own project and own.

search -> inspect -> add -> build

Reuse the capability. Own the source. Pin the version.

Quick start

Install the CLI from a checkout:

git clone https://github.com/MUdevelops/AIRegistry.git
cd AIRegistry
python -m pip install .

Then, in your own project:

ai search rag                       # find components
ai info keyword-rag                 # inspect: releases, files, dependencies
ai add keyword-rag --dry-run        # preview, writes nothing
ai add keyword-rag                  # install the latest release
ai add context-builder@1.1.0        # or pin an exact release
ai list                             # what is installed
ai update                           # move installed components to their latest release
ai remove keyword-rag

Components are copied to ai_components/<name>/ (change it with --dir) and recorded in .airegistry/lock.json. Use one in your code:

import sys
sys.path.insert(0, "ai_components/keyword-rag/src")

from keyword_rag import retrieve

print(retrieve(["Cats purr.", "Python is a language."], "python", top_k=1))

Without pip, run the CLI directly: python ai.py search rag. In a checkout, the CLI reads the local registry.json; otherwise it uses the public registry. Point it elsewhere with ai config registry_url <path-or-https-url>.

Component catalog

Every component is stdlib-only, typed, documented and covered by tests. Dependencies between components are optional and exact; ai add shows them but never installs them automatically. Typical composition:

query-rewriter --> keyword-rag --> context-builder --> prompt-builder
                        \______ retrieval-pipeline ______/
conversation-memory      tool-router --> agent        citation-builder

Browse the full catalog at the live registry.

Registry model

registry/
  keyword-rag/
    versions/
      1.0.0/   component.json  README.md  src/keyword_rag.py
      1.1.0/   component.json  README.md  src/keyword_rag.py

python scripts/build_registry.py validates every release and generates registry.json and website/registry.json; --check fails when they are stale. A release is never changed after it is published: fixes ship as a new version. See ARCHITECTURE.md.

Community publishing

GitHub sign-in -> Submit Component -> Worker validation -> GitHub App -> branch + commit -> Pull Request -> CI + review -> merge

The publisher accepts new components and new releases of existing ones. Nothing is written to main directly. Read CONTRIBUTING.md for the format and rules.

Security

Registry content and submissions are untrusted input. The CLI validates paths and checksums, rejects executable file types and never runs what it installs. See SECURITY.md for details and how to report a vulnerability.

Development

python scripts/build_registry.py --check
python -W error -m unittest discover -s tests -v
python -m py_compile ai.py scripts/build_registry.py
node --check website/app.js
node --check cloudflare/worker.js

Deployment and operations: docs/CLOUD_AUTH.md, docs/MONITORING.md.

License

MIT. See LICENSE.