Get started

Track every AI model your code uses, find out before any of them retire, and migrate cleanly when they do. Install in one line, no account needed.

Install

Three install paths — pick whichever fits your machine. The tab below auto-defaults to your OS.

$ curl -fsSL https://llmstatus.ai/install.sh | bashCopy

After install you have two binaries: mm (short) and llmstatus (descriptive). Same binary — pick whichever you like to type.

5-minute walkthrough

From a fresh install to your first alert. You can stop at step 2 if all you want is the free health check.

$ mm status

LLM Status — registry 20260528T060033Z (today), 380 models
Scanned ./apps/web: 12 reference(s) → 6 model(s), 1 custom

Models in use:
  🔴 retired     openai/gpt-4-0314          retires 2024-06-13   → openai/gpt-4-1  (2)
  🟠 retiring    anthropic/claude-opus-4    retires 2026-06-15   → anthropic/claude-opus-4-7  (1)
  🟢 ok          openai/gpt-5                                                       (3)

⚠ 2 model(s) need attention before they retire.

$ mm fix --dry-run
$ mm fix

$ mm login

$ mm scan

# .github/workflows/mm.yml
- run: npx @modelstatus/cli scan --check

Command reference

Common flags

--api <url> · --key <key> · --project <id|name> ·--yes · --json · --ci · --dry-run ·--sources <list> · --region <r> · --namespace <ns> ·--kube-context <c> · --db <dsn> · --sql-table <t>

How it works

The model lifecycle registry is published as a date-versioned, signed snapshot oncdn.llmstatus.ai. The CLI verifies a mini-TUF trust chain on every fetch and refuses any rollback to an older version:

pinned root key (in the CLI binary)
   → root-signs ─→ keys.json (names the current signing key)
                       → signing-key-signs ─→ latest.json (pointer + sha256)
                                                  → blob: <version>.json (immutable)

Falls back to a local cache when offline (~/.config/llmstatus/registry-cache.json). The signing key can be rotated without shipping a new CLI release. Registry updates auto-publish every 6 hours from sources like models.dev, OpenRouter, and provider lifecycle APIs.

CI checks — block retiring models in pull requests (Pro)

Catch a deprecated or retiring model before it merges. On every push and PR, LLM Status scans the changed files, and fails the check (with an inline annotation on the exact line) if you've introduced a model at or past your threshold. Two ways to turn it on — pick whichever fits your security posture:

Option A — GitHub App (one-click)

Best for fast setup. From your dashboard go to CI runs → Connect GitHub, pick repos, and install. We scan each PR on our side and post a Check Run + a summary comment as the app — no workflow file, no secrets. This is a Pro feature (it runs on our infrastructure and reads your code), and each run is recorded in your CI-drift dashboard.

Option B — GitHub Action (your CI, no code egress)

Best for security-conscious teams: your source never leaves your runners — only resolved model ids + counts are sent (and only if you pass a key). Add .github/workflows/llmstatus.yml:

name: AI model lifecycle
on: [push, pull_request]
jobs:
  llmstatus:
    runs-on: ubuntu-latest
    permissions: { contents: read, pull-requests: write, checks: write }
    steps:
      - uses: actions/checkout@v4
        with: { fetch-depth: 0 }   # so PRs gate only changed files
      - uses: randomartifact/modelmanager/packages/action@v1
        with:
          fail-on: retiring        # retired | retiring | deprecating | none
          api-key: ${{ secrets.LLMSTATUS_API_KEY }}   # omit for the free offline check

The offline check (annotations + the pass/fail gate) is free. Adding your LLMSTATUS_API_KEY(a Pro key) also records each run to your CI-drift dashboard and routes findings to your alert channels.

Free vs Pro

Troubleshooting

macOS: "cannot verify the developer" on first run

The binaries are currently unsigned. The shell installer clears the com.apple.quarantinexattr automatically; if you installed via npm and still see the warning, run xattr -d com.apple.quarantine "$(which mm)". Notarization is a planned follow-up.

mm status says my model is "custom"

The string didn't match any canonical id, slug, or alias in the registry. If it's a real production model we missed, ping it@llmstatus.ai and we'll add it. The registry refreshes every 6 hours, so additions roll out automatically.

Offline / behind a corp proxy

The CLI caches the snapshot at ~/.config/llmstatus/registry-cache.json and falls back to it when fetches fail. Run online once to seed the cache, then it works offline. SetLLMSTATUS_REGISTRY_URL to point at a private mirror if you'd rather not hitcdn.llmstatus.ai from your network.

Questions? it@llmstatus.ai · Source for the CLI is published on npm.