An auto-updating AI Model Catalog available as a Golang Package, CLI Tool, or Server (RESTful, WebSockets, SSE).
____ _
/ ___|| |_ __ _ _ __ _ __ ___ __ _ _ __
\___ \| __/ _` | '__| '_ ` _ \ / _` | '_ \
___) | || (_| | | | | | | | | (_| | |_) |
|____/ \__\__,_|_| |_| |_| |_|\__,_| .__/
|_|
- Why Starmap?
- Key Features
- Installation
- Quick Start
- Architecture
- Core Concepts
- Project Structure
- Choosing Your Approach
- CLI Usage
- Go Package
- Data Sources
- Model Catalog
- HTTP Server
- Configuration
- Development
- Contributing
- API Reference
- License
Building AI applications requires accurate information about models across multiple providers, but:
- Fragmented Information: Each provider has different APIs, documentation formats, and update cycles
- Missing Pricing Data: Many providers don't publish pricing through their APIs
- Rapid Changes: New models launch weekly, capabilities change, prices update
- Integration Complexity: Each provider requires custom code to fetch and parse model data
- No Single Source of Truth: Developers must check multiple sources for complete information
Starmap provides:
- Unified Catalog: Single interface for all AI model information
- Multi-Source Reconciliation: Combines provider APIs with community data for completeness
- Automatic Synchronization: Keep your catalog current with scheduled updates
- Flexible Storage: From in-memory for testing to persistent for production
- Event-Driven Updates: React to model changes in real-time
- Type-Safe Go API: Strongly typed models with comprehensive metadata
- AI Application Developers: Discover and compare models for your use case
- Platform Engineers: Maintain accurate model catalogs for your organization
- Tool Builders: Integrate comprehensive model data into your products
- Researchers: Track model capabilities and pricing trends
- Cost Optimizers: Find the best price/performance for your workloads
✅ Comprehensive Coverage: 500+ models from 10+ providers ✅ Accurate Pricing: Community-verified pricing data via models.dev ✅ Real-time Synchronization: Automatic updates from provider APIs ✅ Flexible Architecture: Simple merging or complex reconciliation ✅ Multiple Interfaces: CLI, Go package, and HTTP Server (REST + WebSocket + SSE) ✅ Production Ready: Thread-safe, well-tested, actively maintained
# Homebrew (macOS/Linux)
brew install agentstation/tap/starmap
# Or install from source
go install github.com/agentstation/starmap/cmd/starmap@latest
# Verify installation
starmap version# Add to your project
go get github.com/agentstation/starmapStarmap provides production-ready container images built with ko using Google's secure Chainguard base images (~2MB, zero CVEs).
Quick Start:
# Pull and run the HTTP server
docker run -p 8080:8080 ghcr.io/agentstation/starmap:latest serve --host 0.0.0.0
# Or use docker-compose (recommended)
docker-compose upUsing Docker Compose:
# 1. Copy environment template
cp .env.example .env
# 2. Edit .env with your API keys (optional)
nano .env
# 3. Start the server
docker-compose up -d
# 4. Check health
curl http://localhost:8080/api/v1/healthAvailable Images:
ghcr.io/agentstation/starmap:latest- Latest stable releaseghcr.io/agentstation/starmap:v0.0.17- Specific versionghcr.io/agentstation/starmap:0.0.17- Specific version (no v prefix)
Supported Platforms:
linux/amd64(x86_64)linux/arm64(ARM 64-bit)
See docs/DOCKER.md for detailed deployment guides including Kubernetes, security hardening, and production best practices.
# List all models
starmap models list
# Filter by provider
starmap models list --provider openai
# Search by capability
starmap models list --capability vision
# Export as JSON
starmap models list --format json > models.jsonpackage main
import (
"fmt"
"log"
"github.com/agentstation/starmap"
)
func main() {
// Create starmap with embedded catalog
sm, err := starmap.New()
if err != nil {
log.Fatal(err)
}
// Get the catalog
catalog, err := sm.Catalog()
if err != nil {
log.Fatal(err)
}
// Find GPT-4 model
model, err := catalog.Model("gpt-4o")
if err == nil {
fmt.Printf("Model: %s\n", model.Name)
fmt.Printf("Context: %d tokens\n", model.ContextWindow)
fmt.Printf("Input Price: $%.2f/1M tokens\n", model.Pricing.Input)
}
}# Set up API keys
export OPENAI_API_KEY=sk-...
export ANTHROPIC_API_KEY=sk-ant-...
# Update catalog from all providers
starmap update
# Update specific provider with auto-approve
starmap update openai -yStarmap uses a layered architecture with clean separation of concerns:
- User Interfaces: CLI, Go package, and HTTP Server (REST + WebSocket + SSE)
- Core System: Catalog management, reconciliation engine, and event hooks
- Data Sources: Provider APIs, models.dev, embedded catalog, and local files
- Storage Backends: Memory, filesystem, embedded, or custom (S3, GCS, etc.)
For detailed architecture diagrams, design principles, and implementation details, see ARCHITECTURE.md.
Starmap's core abstractions provide a clean separation of concerns:
The fundamental abstraction for model data storage and access. Provides CRUD operations, multiple storage backends, and thread-safe collections. See Catalog Package Documentation.
Abstraction for fetching data from external systems (provider APIs, models.dev, local files). Each implements a common interface for consistent data access.
Intelligent multi-source data merging with field-level authority, provenance tracking, and conflict resolution. See Reconciliation Package Documentation.
Comprehensive AI model specification including capabilities (chat, vision, audio), pricing (token costs), limits (context window, rate limits), and metadata. See pkg/catalogs/README.md for the complete Model structure.
For detailed component design and interaction patterns, see ARCHITECTURE.md § System Components.
Starmap follows Go best practices with clear package separation:
pkg/- Public API packages (catalogs, reconciler, sources, errors, etc.)internal/- Internal implementations (providers, embedded data, transport)cmd/starmap/- CLI application
See CONTRIBUTING.md § Project Structure for detailed directory layout and dependency rules.
Starmap provides two levels of data management complexity:
Use Catalog Package (Simple) When:
- ✅ Merging embedded catalog with local overrides
- ✅ Combining two provider responses
- ✅ Testing with mock data
- ✅ Building simple tools
Use Reconciliation Package (Complex) When:
- ✅ Syncing with multiple provider APIs
- ✅ Integrating models.dev for pricing
- ✅ Different sources own different fields
- ✅ Need audit trail of data sources
- ✅ Building production systems
For architecture details and reconciliation strategies, see ARCHITECTURE.md § Reconciliation System.
# Discovery
starmap models list # List all models
starmap providers # List all providers
starmap authors # List all authors
# Model field history
starmap models history gpt-4o # View field provenance
starmap models history gpt-4o --fields=Name # Filter to specific field
starmap models history gpt-4o --fields=Name,ID # Multiple fields
# Update catalog
starmap update # Update all providers
starmap update openai # Update specific provider
starmap update --dry # Preview changes
# Development
starmap validate # Validate configurations
starmap deps check # Check dependency status
starmap completion bash # Generate shell completion# Development: Use file-based catalog
starmap update groq --input-dir ./catalog --dry
# Production: Fresh update with auto-approval
starmap update --force -y
# Custom directories
starmap update --input ./dev --output ./prod
# Specific sources only
starmap update --sources "Provider APIs,models.dev (git)"Some data sources require external tools. Starmap handles missing dependencies gracefully:
# Interactive (default) - Prompts to install or skip
starmap update
# CI/CD - Skip sources with missing dependencies
starmap update --skip-dep-prompts
# Strict mode - Fail if dependencies missing
starmap update --require-all-sources --skip-dep-prompts
# Auto-install - Install dependencies automatically
starmap update --auto-install-depsAvailable Flags:
--auto-install-deps- Automatically install missing dependencies--skip-dep-prompts- Skip sources with missing dependencies without prompting--require-all-sources- Fail if any dependencies are missing (CI/CD mode)
Common Scenario: The models_dev_git source requires bun for building. If missing, Starmap offers to install it or falls back to models_dev_http which provides the same data without dependencies.
Use starmap deps check to verify dependency status before running updates:
# Check all dependencies
starmap deps check
# JSON output for tooling
starmap deps check --format json
# YAML output
starmap deps check --format yamlThe command shows:
- ✅ Available dependencies with version and path
- ❌ Missing dependencies with installation instructions
- ℹ️ Sources that don't require any dependencies
Example output:
Dependency Status:
┌────────────────────────────┬────────────────────────┬──────────────────┬─────────┬───────────────────────┐
│ SOURCE │ DEPENDENCY │ STATUS │ VERSION │ PATH │
├────────────────────────────┼────────────────────────┼──────────────────┼─────────┼───────────────────────┤
│ local_catalog (optional) │ - │ ✅ None required │ - │ - │
│ providers │ - │ ✅ None required │ - │ - │
│ models_dev_git (optional) │ Bun JavaScript runtime │ ✅ Available │ 1.2.21 │ /opt/homebrew/bin/bun │
│ │ Git version control │ ✅ Available │ 2.51.0 │ /opt/homebrew/bin/git │
│ models_dev_http (optional) │ - │ ✅ None required │ - │ - │
└────────────────────────────┴────────────────────────┴──────────────────┴─────────┴───────────────────────┘
Additional Information:
Bun JavaScript runtime (models_dev_git):
Description: Fast JavaScript runtime for building models.dev data
Why needed: Builds api.json from models.dev TypeScript source
Summary:
┌────────────────────────────────┬───────┐
│ STATUS │ COUNT │
├────────────────────────────────┼───────┤
│ ✅ Available │ 2 │
│ ℹ️ Sources without dependencies │ 3 │
└────────────────────────────────┴───────┘
✅ All required dependencies are available.
# Required for provider syncing
export OPENAI_API_KEY=sk-...
export ANTHROPIC_API_KEY=sk-ant-...
export GOOGLE_API_KEY=...
export GROQ_API_KEY=...
# Optional for Google Vertex
export GOOGLE_VERTEX_PROJECT=my-project
export GOOGLE_VERTEX_LOCATION=us-central1import (
"github.com/agentstation/starmap"
"github.com/agentstation/starmap/pkg/catalogs"
"github.com/agentstation/starmap/pkg/reconciler"
)// Default embedded catalog with auto-updates
sm, _ := starmap.New()
catalog, _ := sm.Catalog()
// Query models
model, _ := catalog.Model("claude-3-opus")
fmt.Printf("Context: %d tokens\n", model.ContextWindow)// React to catalog changes
sm.OnModelAdded(func(model catalogs.Model) {
log.Printf("New model: %s", model.ID)
})
sm.OnModelUpdated(func(old, new catalogs.Model) {
if old.Pricing.Input != new.Pricing.Input {
log.Printf("Price changed for %s", new.ID)
}
})// Use filesystem for development
catalog, _ := catalogs.New(
catalogs.WithPath("./my-catalog"),
)
sm, _ := starmap.New(
starmap.WithInitialCatalog(catalog),
)// Sync with all configured provider APIs
result, err := sm.Sync(ctx,
sync.WithProviders("openai", "anthropic"),
sync.WithDryRun(false),
)
if err != nil {
log.Fatal(err)
}
fmt.Printf("Added: %d models\n", result.Added)
fmt.Printf("Updated: %d models\n", result.Updated)
fmt.Printf("Removed: %d models\n", result.Removed)updateFunc := func(current catalogs.Catalog) (catalogs.Catalog, error) {
// Custom sync logic
// Could call provider APIs, merge data, etc.
return updatedCatalog, nil
}
sm, _ := starmap.New(
starmap.WithAutoUpdateInterval(30 * time.Minute),
starmap.WithUpdateFunc(updateFunc),
)// Find vision-capable models under $10/M tokens
models := catalog.Models()
models.ForEach(func(id string, model *catalogs.Model) bool {
if model.Features.Vision && model.Pricing.Input < 10 {
fmt.Printf("Vision model: %s ($%.2f/M)\n",
model.ID, model.Pricing.Input)
}
return true
})Starmap combines data from multiple sources:
- Provider APIs: Real-time model availability (OpenAI, Anthropic, Google, etc.)
- models.dev: Community-verified pricing and metadata (models.dev)
- Embedded Catalog: Baseline data shipped with starmap
- Local Files: User customizations and overrides
For detailed source hierarchy, authority rules, and how sources work together, see ARCHITECTURE.md § Data Sources.
Starmap includes 500+ models from 10+ providers (OpenAI, Anthropic, Google, Groq, DeepSeek, Cerebras, and more). Each package includes comprehensive documentation in its README.
Start a production-ready REST API server for programmatic catalog access:
# Start on default port 8080
starmap serve
# Custom configuration
starmap serve --port 3000 --cors --auth --rate-limit 100
# With specific CORS origins
starmap serve --cors-origins "https://example.com,https://app.example.com"Features:
- RESTful API: Models, providers, search endpoints with filtering
- Real-time Updates: WebSocket (
/api/v1/updates/ws) and SSE (/api/v1/updates/stream) - Performance: In-memory caching, rate limiting (per-IP)
- Security: Optional API key authentication, CORS support
- Monitoring: Health checks (
/health,/api/v1/ready), metrics endpoint - Documentation: OpenAPI 3.1 specs at
/api/v1/openapi.json
API Endpoints:
# Models
GET /api/v1/models # List with filtering
GET /api/v1/models/{id} # Get specific model
POST /api/v1/models/search # Advanced search
# Providers
GET /api/v1/providers # List providers
GET /api/v1/providers/{id} # Get specific provider
GET /api/v1/providers/{id}/models # Get provider's models
# Admin
POST /api/v1/update # Trigger catalog sync
GET /api/v1/stats # Catalog statistics
# Health
GET /health # Liveness probe
GET /api/v1/ready # Readiness checkConfiguration Flags:
--port: Server port (default: 8080)--host: Bind address (default: localhost)--cors: Enable CORS for all origins--cors-origins: Specific CORS origins (comma-separated)--auth: Enable API key authentication--rate-limit: Requests per minute per IP (default: 100)--cache-ttl: Cache TTL in seconds (default: 300)
Environment Variables:
HTTP_PORT=8080
HTTP_HOST=0.0.0.0
STARMAP_API_KEY=your-api-key # If --auth enabledFor full server documentation, see internal/server/README.md.
# Provider API Keys
OPENAI_API_KEY=sk-...
ANTHROPIC_API_KEY=sk-ant-...
GOOGLE_API_KEY=...
GROQ_API_KEY=...
DEEPSEEK_API_KEY=...
CEREBRAS_API_KEY=...
# Google Vertex (optional)
GOOGLE_VERTEX_PROJECT=my-project
GOOGLE_VERTEX_LOCATION=us-central1
GOOGLE_APPLICATION_CREDENTIALS=/path/to/service-account.json
# Starmap Configuration
STARMAP_CONFIG=/path/to/config.yaml
STARMAP_CACHE_DIR=/var/cache/starmap
STARMAP_LOG_LEVEL=infoCheck and verify your authentication setup:
# Check authentication status for all providers
starmap providers
# Test credentials by making test API calls
starmap providers --test
# Test specific provider
starmap providers openai --test
# JSON output for automation
starmap providers --output json
# Manage Google Cloud authentication
starmap auth gcloudThe providers command shows:
- Which providers have configured credentials
- Authentication method (API key, ADC, OAuth)
- Credential source (environment variable, config file, application default)
- Missing credentials with setup instructions
- Provider details (name, ID, location, type, models count)
# ~/.starmap.yaml
providers:
openai:
api_key: ${OPENAI_API_KEY}
rate_limit: 100
catalog:
type: embedded
auto_update: true
update_interval: 1h
sync:
sources:
- Provider APIs
- models.dev (git)
auto_approve: false
logging:
level: info
format: jsonTo contribute or develop locally:
git clone https://github.com/agentstation/starmap.git
cd starmap
make allSee CONTRIBUTING.md for complete development setup, testing guidelines, and contribution process.
We welcome contributions! Please see CONTRIBUTING.md for:
- Development setup and workflow
- How to add new providers
- Testing requirements
- Pull request process
- Code guidelines
Quick links:
This project is licensed under the GNU Affero General Public License v3.0 (AGPL-3.0).
The AGPL ensures that:
- Source code remains open for any network use
- Modifications must be shared with users
- The community benefits from all improvements
See LICENSE file for full details.
For complete API documentation including all types, interfaces, and functions, see API.md.
Quick links: