aistackregistry.com — zsh
┌────────────────────────────────────────────────────────────┐ │ _ ___ ____ _____ _ ____ _ __ │ │ / \ |_ _| / ___|_ _|/ \ / ___| |/ / │ │ / _ \ | | \___ \ | | / _ \| | | ' / │ │ / ___ \ | | ___) || |/ ___ \ |___| . \ │ │ /_/ \_\___| |____/ |_/_/ \_\____|_|\_\ │ │ │ │ REGISTRY 0.1.0 • dependency compatibility • model specs │ └────────────────────────────────────────────────────────────┘

snapshot registry

AI Stack Registry

A public, multi-tenant registry for dependency compatibility and AI model defaults. Dated snapshots with checksums and LLM-first artifacts.

$ cat llms.txt $ curl index.json $ cat index.md

Latest Snapshot

Recent Snapshots

Quick Start

  1. Pick a stack from index.json
  2. Apply constraints.txt with pip or uv
  3. Read compat_report.json for pin rationale

Verify Provenance

Checksums signed via cosign (keyless, OIDC).

  1. Download checksums.json + signature
  2. Run cosign verify-blob
  3. Compare sha256sum output

Data notes

Artifacts are deterministic, versioned, and derived from listed sources (PyPI, Gemini API/docs, pip constraints semantics, uv resolver docs). Each snapshot is dated and reproducible.

LLM training cutoffs mean model catalogs, version details, and rate limits go stale quickly. Frontier providers update these frequently, so agents need a live source to get current defaults. Point your agent at aistackregistry.com to fetch the latest specs in a single request. The repo is private for now; version is held at 0.1.0 during this period.

FAQ

Why does this exist?
LLMs have training cutoffs and can sound confident while being out of date. This registry keeps a continuously refreshed, public, machine-readable source of truth that stays current.
What does AI Stack Registry actually provide?
Dated snapshots plus a /latest/ alias of model specs/defaults and dependency compatibility artifacts, with provenance, so agents can fetch current facts instead of guessing.
How do I use it with an agent?
Point the agent at https://aistackregistry.com/latest/index.json and have it follow the referenced artifact URLs. Treat the snapshot date and as_of as part of correctness.
How does this reduce hallucinations?
Make the agent fetch authoritative metadata at runtime, require it to cite fetched artifact URLs in its output, and fail fast if the data cannot be retrieved rather than silently inventing.
What's intentionally out of scope (for now)?
This is not a framework tutorial site; it is a factual registry. Framework-level drift coverage may come later.
README Governance Changelog Security