@ifi/pi-provider-ollama

Experimental Ollama provider package for pi with both local and cloud support.

What it does

• Registers a local ollama provider via pi.registerProvider(...)
• Detects whether the Ollama CLI is available on Windows and Unix-like systems
• Auto-discovers installed local Ollama models from the running daemon
• Surfaces downloadable local candidates from the cloud catalog with context-window metadata
• Adds /login ollama-cloud support using an Ollama API key flow
• Discovers the current Ollama Cloud model catalog even before login, then stores refreshed metadata with the login credential
• Exposes local models in /model as ollama/<model-id>
• Exposes cloud models in /model as ollama-cloud/<model-id>
• Prompts to download a missing local model when you select ollama/<model-id> and uses the Ollama CLI to pull it
• Adds /ollama:status|refresh-models|models|info|pull for a unified local + cloud workflow

Install

pi install npm:@ifi/pi-provider-ollama

This package is intentionally separate from @ifi/oh-pi for now.

Use

Local Ollama

1. Install the package
2. Make sure the ollama CLI is installed and a local Ollama instance is running
3. Open /model and select an ollama/... model
4. If the model is not installed yet, pi will prompt to download it with the Ollama CLI
5. Run /ollama:refresh-models whenever you pull or remove local models outside pi

Ollama Cloud

1. Install the package
2. Open /model or run /ollama:models to browse the public Ollama Cloud catalog
3. Run /login ollama-cloud before using an ollama-cloud/... model
4. Create an API key on Ollama when pi opens the keys page
5. Paste the key back into pi
6. Open /model and select an ollama-cloud/... model
7. Optionally run /ollama:refresh-models later to refresh both local and cloud catalogs

Commands

• /ollama:status — show local CLI + daemon status, local install/download counts, and cloud auth/catalog status
• /ollama:refresh-models — refresh both local and cloud Ollama models
• /ollama:models — list local and cloud Ollama models with source/capability badges for easier selection
• /ollama:info <model> — show detailed metadata for a local or cloud Ollama model, including context window size
• /ollama:pull <model> — manually download a local Ollama model via the Ollama CLI
• /ollama-cloud status — backward-compatible cloud-only status alias
• /ollama-cloud refresh-models — backward-compatible cloud-only refresh alias

Notes

• Local Ollama uses the daemon's OpenAI-compatible /v1 API plus /api/show metadata.
• If the Ollama CLI is missing, pi warns that only ollama-cloud/... models are available because Ollama is not installed locally.
• pi prefers the broader cloud catalog for browsing and uses the cloud metadata to annotate local download candidates with accurate context sizes.
• Cloud Ollama always discovers the public catalog first, then merges any authenticated metadata so login cannot shrink the visible model list.
• Ollama's documented API-key flow is still used for actual cloud access.
• Local model discovery is dynamic and installation-specific, so there is no static local fallback catalog.
• Cloud model discovery falls back to a bundled cloud catalog when live discovery is unavailable.
• Costs are currently left at zero because Ollama does not expose stable per-token pricing for local or cloud use in a way that pi can rely on here.

Test hooks

These environment variables exist mainly for tests and local debugging:

Local

• PI_OLLAMA_LOCAL_API_URL
• PI_OLLAMA_LOCAL_MODELS_URL
• PI_OLLAMA_LOCAL_SHOW_URL
• PI_OLLAMA_LOCAL_ORIGIN
• OLLAMA_HOST

Cloud

• PI_OLLAMA_CLOUD_API_URL
• PI_OLLAMA_CLOUD_MODELS_URL
• PI_OLLAMA_CLOUD_SHOW_URL
• PI_OLLAMA_CLOUD_KEYS_URL
• PI_OLLAMA_CLOUD_ORIGIN
• OLLAMA_HOST_CLOUD
• OLLAMA_API_KEY

Legacy OLLAMA_CLOUD_* env names are also accepted for cloud compatibility with earlier iterations of this package.