modelgen

A small Go CLI that generates AI model configuration files for the Raycast AI extensions and the Zed editor from an LLM gateway.

The tool reads a gateway's /v1/models endpoint, fetches metadata from each upstream source, and emits a provider configuration for Raycast and Zed with model display names suffixed by the upstream provider name.

Outputs

• Raycast: providers.yaml for Raycast AI.
• Zed: a settings.json snippet for language_models.openai_compatible.

Installation

Pre-built binary (macOS)

curl -fsSL https://code.paulg.it/paulgit/modelgen/raw/branch/main/install.sh | sh

The script detects your OS and architecture, downloads the matching binary from the latest release, verifies it against the published SHA256SUMS, and installs it to ~/.local/bin. If ~/.local/bin is not already on your PATH, the script prints a one-line command to add it.

To review the script before running it:

curl -fsSL https://code.paulg.it/paulgit/modelgen/raw/branch/main/install.sh | less

To install to a custom directory:

curl -fsSL https://code.paulg.it/paulgit/modelgen/raw/branch/main/install.sh | sh -s -- --prefix /opt/bin

Build from source

go build -o modelgen .
# or install via go install if the repo is published

Configuration

Copy config.example.toml to ~/.config/modelgen/config.toml and adjust as needed.

[general]
generators = ["raycast", "zed"]

[raycast]
base_url = "https://ai.paulg.it/v1"
api_key = "pg-..."
allow_upstream = ["openrouter", "ollama", "openai"]
allow_patterns = [".*"]
exclude_patterns = []

[zed]
base_url = "https://ai.paulg.it/v1"
api_key = "pg-..."
allow_upstream = ["openrouter", "ollama", "openai"]
allow_patterns = [".*"]
exclude_patterns = []

Each target (raycast and zed) has its own API key, gateway URL, and filtering options. API keys can also be provided via the RAYCAST_API_KEY and ZED_API_KEY environment variables. The Raycast API key is written directly into the generated providers.yaml. Zed does not read API keys from settings.json — configure the key in Zed's UI or keychain instead.

1Password Integration

API keys can optionally be retrieved from 1Password using the op CLI, keeping secrets out of config files and environment variables.

Setup

1. Install the 1Password CLI.
2. Sign in: op signin
3. Store your API key as a 1Password item (e.g. a "Secure Note" or "API Key" item).
4. Reference the item in your config using an op:// URI in the api_key field:

[raycast]
base_url = "https://ai.paulg.it/v1"
api_key = "op://Personal/modelgen/raycast-api-key"

[zed]
base_url = "https://ai.paulg.it/v1"
api_key = "op://Personal/modelgen/zed-api-key"

You can also use op:// references in environment variables:

RAYCAST_API_KEY="op://Personal/modelgen/raycast-api-key" ./modelgen

Resolution Order

API keys are resolved in the following order (first match wins):

1. api_key in the config file (if value starts with op://, resolved via 1Password)
2. <TARGET>_API_KEY environment variable (if value starts with op://, resolved via 1Password)
3. Error: no API key found

This means you can use 1Password for local development and literal keys for CI/CD without changing your config.

Usage

# Generate configured outputs
./modelgen

# Dry-run: print outputs to stdout instead of writing files
./modelgen --dry-run

# Merge generated Zed config into an existing settings.json
./modelgen --merge-zed ~/Library/Application\ Support/Zed/settings.json

# Run only the Zed generator
./modelgen --generators zed

# Use a custom config file
./modelgen --config /path/to/config.toml

# Print version
./modelgen --version

# Clear the on-disk metadata cache (forces a re-fetch on next run)
./modelgen --clear-cache

How It Works

1. Fetches the list of available models from the configured gateway /v1/models endpoint for each enabled generator.
2. Parses the upstream provider prefix from each model ID, e.g.:
   • openrouter/anthropic/claude-sonnet-4-5
   • ollama/llama3.1:latest
3. Enriches each model from the appropriate upstream metadata source:
   • OpenRouter models use OpenRouter's /v1/model/{id} endpoint.
   • Ollama models use Ollama Cloud's /api/tags endpoint.
   • Other upstreams fall back to models.dev metadata, then to sensible defaults.
4. Caches metadata locally in ~/.cache/modelgen/metadata-cache-<target>.json with a default TTL of 30 days.
5. Generates a provider named Paul Git with model display names such as Claude Sonnet 4.5 (OpenRouter).

Project Layout

modelgen/
  cmd/root.go              # CLI orchestration
  internal/
    cache/                 # On-disk metadata cache
    config/                # TOML config loading
    fetch/                 # Gateway fetcher and upstream enrichment
    generate/              # Raycast and Zed output generators
    secret/                # Pluggable secret providers (1Password, etc.)
    types/                 # Canonical model/provider types
  .forgejo/workflows/      # Forgejo Actions CI and release workflows
  config.example.toml
  install.sh               # One-line macOS install script (curl | sh)
  Makefile

Development

make test        # run tests
make vet         # run go vet
make build       # build the binary
make dry-run     # run with --dry-run

Security Notes

• The Raycast fetch API key is written directly into the generated providers.yaml. Zed does not read API keys from settings.json — configure the key in Zed's UI or keychain instead.
• Secret-bearing output files are written with restrictive permissions.
• API keys can be stored in 1Password and retrieved at runtime by setting api_key to an op:// reference. This works for both config file values and environment variables.
• See config.example.toml for the recommended configuration style.