══════════════════════════════════════════════════════════════
══════════════════════════════════════════════════════════════


 ███╗   ███╗ ██████╗ ██████╗ ███████╗██╗     ██████╗ ██████╗
 ████╗ ████║██╔═══██╗██╔══██╗██╔════╝██║     ██╔══██╗██╔══██╗
 ██╔████╔██║██║   ██║██║  ██║█████╗  ██║     ██║  ██║██████╔╝
 ██║╚██╔╝██║██║   ██║██║  ██║██╔══╝  ██║     ██║  ██║██╔══██╗
 ██║ ╚═╝ ██║╚██████╔╝██████╔╝███████╗███████╗██████╔╝██████╔╝
 ╚═╝     ╚═╝ ╚═════╝ ╚═════╝ ╚══════╝╚══════╝╚═════╝ ╚═════╝


══════════════════════════════════════════════════════════════
══════════════════════════════════════════════════════════════

REST API service for AI model metadata and costs.

Free hosted API: modeldb.axiom.co

Data source: LiteLLM's model database

Description

ModelDB provides a comprehensive database of AI language models with their associated metadata including costs, context windows, capabilities, and provider information. Designed for programmatic access by applications or data workloads requiring model selection, cost estimation, or capability comparison.

Features

• Built from LiteLLM's community-maintained model pricing data
• 1000+ models from 40+ providers
• Real-time pricing in both per-token and per-million formats
• Model capabilities (vision, function calling, JSON mode)
• Context window specifications
• CSV export support
• Field projection for bandwidth optimization
• OpenAPI 3.1 specification
• CORS-enabled for browser usage
• No authentication required

Quick Start

# List all models
curl https://modeldb.axiom.co/api/v1/models

# Get specific model
curl https://modeldb.axiom.co/api/v1/models/gpt-4o

# Filter by provider
curl "https://modeldb.axiom.co/api/v1/models?providers=openai,anthropic"

# Filter by capabilities
curl "https://modeldb.axiom.co/api/v1/models?capability=vision,function_calling"

# Export as CSV
curl "https://modeldb.axiom.co/api/v1/models?format=csv" > models.csv

# Export as JSON Lines (JSONL)
curl "https://modeldb.axiom.co/api/v1/models?format=jsonl" > models.jsonl

# Project specific fields
curl "https://modeldb.axiom.co/api/v1/models?project=model_id,input_cost_per_million"

# Pretty print JSON output
curl "https://modeldb.axiom.co/api/v1/models/gpt-4o?pretty"

API Routes

Route	Description
GET /api/v1/models	List all models with filtering
GET /api/v1/models/:id	Get specific model by ID
GET /api/v1/providers	List all providers
GET /api/v1/providers/:id	Get models for a provider
GET /api/v1/metadata	Get sync metadata
GET /openapi.json	OpenAPI specification
GET /ui	Interactive API documentation

Query Parameters

Models Endpoint

• prefixes - Comma-separated list of prefixes to filter models
• providers - Comma-separated list of providers to filter models
• type - Comma-separated list of model types to filter
• capability - Comma-separated list of capabilities to filter (function_calling, vision, json_mode, parallel_functions)
• deprecated - Filter models by deprecation status (true/false)
• project - Comma-separated fields to return
• format - Output format (json, csv, or jsonl)
• headers - Include headers in CSV output (defaults to false, use headers=true)
• fill-with-zeros - Replace null values with 0
• pretty - Pretty print JSON output with indentation (presence of parameter enables it)

Installation

# Clone repository
git clone https://github.com/axiomhq/modeldb.git
cd modeldb

# Install dependencies
npm install

# Start development server
npm run dev

# Run tests
npm test

# Update model data from LiteLLM
npm run sync

Deployment

# Deploy to Cloudflare Workers
npm run deploy

# Deploy to production
npm run deploy:prod

Data Refresh (Cron + KV)

Model data now refreshes hourly via a Cloudflare Cron and is stored in KV (and warmed into Workers Cache). To force a refresh (e.g., first deploy), call the admin endpoint:

curl -X POST \
  -H "Authorization: Bearer $ADMIN_TOKEN" \
  https://modeldb.axiom.co/admin/refresh?force=true

Configure KV binding MODELS_KV and set ADMIN_TOKEN using Wrangler secrets. The service reads all responses from the warmed cache/KV; no runtime fetches are performed for public requests.

Admin endpoints (token required):

# Force refresh now (skip ETag with force=true)
curl -X POST -H "Authorization: Bearer $ADMIN_TOKEN" \
  https://modeldb.axiom.co/admin/refresh?force=true

# Health/manifest and cache status
curl -H "Authorization: Bearer $ADMIN_TOKEN" \
  https://modeldb.axiom.co/admin/health

Architecture

ModelDB uses scheduled refresh + cache-first serving:

1. Hourly cron fetches LiteLLM JSON and transforms it to prebuilt artifacts (list, map, providers, metadata).
2. Artifacts are written to Cloudflare KV and warmed into Workers Cache under /cache/v1/latest/*.json.
3. API routes read from Workers Cache, falling back to KV, then to bundled src/data/* as a last resort.
4. Admin refresh can force regeneration; a health endpoint reports manifest and cache status.

Contributing

See CONTRIBUTING.md for guidelines.

Model Data Updates

Model information is maintained by the LiteLLM community and synced hourly. To propose data changes:

• Submit PRs to LiteLLM's model database
• This service ingests updates automatically via cron

Acknowledgments

• LiteLLM and community for maintaining model data

License

MIT © Axiom, Inc.