pi-ollama-cloud

An Ollama Cloud provider extension for the Pi coding agent. It connects directly to Ollama Cloud, so no local Ollama server is required.

Features

• Automatically discovers available Ollama Cloud chat models and refreshes Pi's model catalog.
• Adds ollama_web_search and ollama_web_fetch tools using Ollama Cloud's web API.
• Optionally displays Ollama Cloud usage in the footer and provides a usage command.
• Reports comparable per-token costs in Pi. Ollama Cloud is subscription/credit billed, so /cost values are estimates, not actual charges.

Requirements

• Pi 0.84.0 or later.
• An Ollama Cloud API key.

Install

pi install npm:pi-ollama-cloud

To try it without installing:

pi -e npm:pi-ollama-cloud

Other install options are available in the Pi extension docs.

Setup

In Pi, run /login, choose Use an API key, select Ollama Cloud, and paste your key. Pi stores it in ~/.pi/agent/auth.json. Alternatively, set OLLAMA_API_KEY in the environment before starting Pi.

Select a model with /model or Ctrl+L. Models refresh automatically on startup, when /model opens, and with pi update --models. A credential must be configured for Pi to run the live refresh. Only models suitable for chat and advertising tool support are offered.

Optional settings

Settings can be placed in ~/.pi/agent/ollama-cloud.json (user-wide) or .pi/ollama-cloud.json (project-local, takes precedence).

{
  "webTools": true,
  "usageStatus": false
}

PI_OLLAMA_WEB_TOOLS=0 (also false, no, or off) disables web tools regardless of config.

Tools and commands

The web tools use the configured Ollama Cloud key directly. Search returns snippets and can expand cached results; fetch returns page text in chunks. Results are cached on disk for 24 hours. Use refresh=true to bypass the cache. Cache location, size limits, and paging options are documented under Web tool details.

The usage bar is off by default. It refreshes at most every five minutes and is cleared when switching away from Ollama Cloud. See Usage details for its behavior and the exported API for custom status bars.

Model and cost notes

Pi refreshes the live model catalog and stores it in its own model store. The generated model list included with the extension is only a first-launch fallback. Models that use a different protocol, such as Ollama's Clef decision models served through /v1/systemone, are excluded from the chat model picker.

Pi's /cost uses the published rates from ollama.com/pricing as comparable per-token estimates. They are not actual subscription charges. Models without a published price report zero cost. The completion API does not report cached-token usage, so cost estimates cannot account for cache hits.

Troubleshooting

• No models appear: configure an API key and reopen /model. Pi only runs the live catalog refresh when credentials are available.
• A refresh is incomplete: individual model detail requests can time out; successful models are retained. Retry later if the desired model is missing.
• DeepSeek tool calls show raw markup: some DeepSeek responses leak native tool-call markup through the OpenAI-compatible endpoint. Retry or switch models.

For changes, tests, live smoke tests, and release instructions, see the development guide and release guide.

Difference from ollama launch pi

ollama launch pi configures Pi to use a local Ollama server, which can proxy cloud models. This extension connects directly to ollama.com and does not require a local server. The two providers can be used together: local Ollama models appear under ollama, while this extension's cloud models appear under ollama-cloud.