LLM Registry

A lightweight, cache-optimized LLM registry that auto-updates from Artificial Analysis API daily and runs on Cloudflare Workers.

Quick Start

Deploy the API

1. Click the Deploy button above

   Build command: None

   Deploy command: bunx wrangler deploy

   Root directory: /

2. Create a KV namespace

   Go to Storage & Databases > Workers KV

   Copy the Namespace ID

3. Change wrangler.toml

   Use your custom domain or delete that line to use your-subdomain.workers.dev

   Change the kv_namespace id to your created Namespace ID

4. Set Variables and Secrets from .env

   See Environment Variables below

5. If you want to deploy it manually without waiting until 2AM

   bunx wrangler deploy

   bun run seed

   wrangler kv key put --binding REGISTRY manifest --path ./data/manifest-preview.json

Use the Client Library

bun add @rttnd/llm

import { createRegistry } from '@rttnd/llm'

const registry = createRegistry({
  baseUrl: 'https://llm.your-subdomain.workers.dev',
})

// Get all models
const { data: models } = await registry.getModels()

// Search for models
const { data: visionModels } = await registry.searchModels({
  capability: 'vision',
  minIq: 3,
})

// Get specific model details
const { data: model } = await registry.getModel('openai', 'gpt-5')

See the Client Library Documentation for more details.

Direct KV access

When your app already runs on Cloudflare Workers you can skip HTTP requests entirely and read the manifest straight from the shared KV namespace:

import { createKVRegistry } from '@rttnd/llm'

export default {
  async fetch(request, env) {
    const registry = createKVRegistry({ kv: env.REGISTRY })
    const { data: models, error } = await registry.getModels()
    if (error)
      throw error

    return Response.json(models)
  },
}

• Works inside Server Functions with meta frameworks that use Nitro.
• Override manifestKey if you store multiple manifests in the same namespace.
• KV namespace IDs are safe to keep in your repo; they only work inside accounts that bind them via Cloudflare dashboard.

Custom providers and models

The registry includes some official overrides (e.g. Azure, Groq, Cerebras) on the server side, but you can also add your own providers and models in your app.

• Server-side (this repo): official overrides live in packages/server/src/custom and are merged into the manifest during the transform step.
• Client-side (your app): you can take the manifest and apply your own overrides at runtime using the shared helper.

Example: add OpenRouter as a custom provider, and a couple of OpenRouter models, on the client:

import type { ModelOverride, ProviderOverride } from '@rttnd/llm'
import { applyOverrides, createRegistry } from '@rttnd/llm'

const registry = createRegistry({ baseUrl: 'https://llm.your-subdomain.workers.dev' })

const openRouterProvider: ProviderOverride = {
  value: 'openrouter',
  name: 'OpenRouter',
  website: 'https://openrouter.ai/',
  status: 'latest',
}

const openRouterModels: ModelOverride[] = [
  {
    provider: 'openrouter',
    value: 'openai/gpt-5-pro',
    inheritFrom: { provider: 'openai', value: 'gpt-5-1' }, // reuse AA metrics/capabilities
  },
  {
    provider: 'openrouter',
    value: 'openai/gpt-oss-20b:free',
    inheritFrom: { provider: 'openai', value: 'gpt-oss-20b' },
    pricing: { input: 0, output: 0, blended: 0 }, // mark as free on OpenRouter
  },
]

const { data: manifest } = await registry.getManifest()
const { providers, models } = applyOverrides(manifest.providers, manifest.models, {
  providers: [openRouterProvider],
  models: openRouterModels,
})

API Endpoints

GET /v1/manifest

Returns the complete manifest with all providers and models.

Response:

{
  "version": "v1.5c3e2c7d9f8a",
  "etag": "\"5c3e2c7d9f8a6b4c1d0e9f8a6b4c1d0e\"",
  "generatedAt": "2025-01-15T02:00:00.000Z",
  "providers": [/* ... */],
  "models": [/* ... */]
}

GET /v1/providers

Returns list of all providers.

Response:

[
  {
    "value": "openai",
    "name": "OpenAI",
    "keyPlaceholder": "sk-...",
    "website": "https://platform.openai.com/api-keys",
    "status": "latest"
  }
]

GET /v1/providers/:providerId/models

Returns models for a specific provider.

Response:

[
  {
    "id": "...",
    "value": "gpt-5",
    "provider": "openai",
    "name": "GPT-5",
    "alias": "GPT-5",
    "iq": 5,
    "speed": 3,
    "pricing": {
      "input": 1.25,
      "output": 10,
      "blended": 3.44
    }
  }
]

GET /v1/models/search

Search models with query parameters rather than downloading the full manifest.

Query Parameters:

• name - Filter by partial match across name, value, or alias. Accepts a single string or an array of strings
• provider - Restrict results to one or more provider slugs
• capability - Require a capability (text, vision, reasoning, toolUse, json, audio)
• minIq - Minimum IQ score (0-5)
• minSpeed - Minimum speed score (0-5)
• status - Filter by one or more statuses (latest, preview, all)

Response:

[
  {
    "id": "...",
    "value": "gpt-5",
    "provider": "openai",
    "name": "GPT-5",
    "iq": 5,
    "speed": 3,
    "capabilities": {
      "vision": true,
      "text": true
    }
  }
]

GET /v1/version

Returns version info for update checks.

Response:

{
  "version": "v1.5c3e2c7d9f8a",
  "etag": "\"5c3e2c7d9f8a6b4c1d0e9f8a6b4c1d0e\"",
  "generatedAt": "2025-01-15T02:00:00.000Z"
}

GET /health

Health check endpoint.

Response:

{
  "status": "ok"
}

Configuration

Environment Variables

Set these in Cloudflare dashboard or via wrangler secret:

• AA_API_KEY - Your Artificial Analysis API key (https://artificialanalysis.ai/documentation)
• ALLOWED_ORIGINS - Comma-separated CORS origins (e.g., https://app.com,https://*.example.com), use * to allow all origins.

Cron Schedule

The default cron schedule is daily at 2 AM UTC in wrangler.toml:

[triggers]
crons = [ "0 2 * * *" ]

Architecture

Data Flow

1. Cloudflare Cron fetches from Artificial Analysis API daily
2. Transform converts AA data to normalized model format
3. Store saves manifest to Workers KV
4. Serve manifest from Worker
5. GitHub Action publishes human-readable snapshot
6. Me: I add the missing data to the registry that's missing from the AA API

Scoring System

IQ Score (0-5):

• 5: Intelligence Index ≥ 65
• 4: Intelligence Index ≥ 55
• 3: Intelligence Index ≥ 45
• 2: Intelligence Index ≥ 35
• 1: Intelligence Index ≥ 25
• 0: < 25

Speed Score (0-5):

• 5: ≥ 300 tokens/sec
• 4: ≥ 200 tokens/sec
• 3: ≥ 100 tokens/sec
• 2: ≥ 50 tokens/sec
• 1: ≥ 25 tokens/sec
• 0: < 25

License

MIT

Credits

Data from Artificial Analysis