Ollama (Local Models)

Gemini Scribe can route chat, summary, completions, rewrite, and agent tool-calling through a local Ollama daemon instead of the Google Gemini API. Use this when you want offline operation, full data privacy, or to avoid API quota limits.

Setup

1. Install Ollama — Download the installer from ollama.com and run it. The daemon listens on http://localhost:11434 by default.
2. Pull a model — In a terminal, fetch any model that supports tool calling:

   ollama pull llama3.2
   ollama pull qwen2.5:7b
   ollama pull llava:13b      # for vision (image input)

3. Switch the provider in Gemini Scribe — Open Settings → Gemini Scribe → Provider and choose Ollama (local).
4. Pick a model — The Chat / Summary / Completion dropdowns now list whatever you have pulled. In Settings → General, click Refresh in the Refresh model list row if a new pull doesn't show up.

If the daemon runs on a different host or port, edit the Ollama Base URL field (e.g. http://10.0.0.5:11434).

What works

• Agent chat with streaming, tool calling, and conversation memory
• Drag-and-drop / paste of image attachments to vision models (e.g. llava, moondream, qwen2.5-vl)
• File summarization, IDE-style completions, selection rewriting
• Custom prompts, projects, agent skills, scheduled tasks, MCP servers

What does not work in Phase 1

These features depend on Gemini built-in services and are hidden when Ollama is the active provider:

Feature	Why it's gated	Workaround
Google Search tool	Uses Gemini's googleSearch grounding	Switch to Gemini for search-heavy sessions
URL Context (web fetch)	Uses Gemini's URL Context API	Paste content into a note, then read_file
Deep Research	Built on Gemini multi-step search	Switch to Gemini
Image generation	Ollama has no image-generation API	Switch to Gemini for image-gen
RAG / Vault Search	Uses Gemini File Search Store	Future phase: Ollama embeddings
PDF / audio / video attachments	Ollama only accepts images	Convert to image or text first

Switching back to Gemini at any time restores all features — settings persist across switches.

Tips

• Tool calling — Most modern instruct models support function calling; older or very small models may not. If the agent loop stalls, try a different model (Llama 3.2, Qwen 2.5, Mistral 0.3 are good starting points).
• Context window — Local models often have smaller context than Gemini. Compaction triggers at the percentage set by Context Compaction Threshold (default 20%) of an estimated 32k-token window; long sessions will summarise older turns earlier than they do on Gemini.
• Token counts — Ollama does not expose a countTokens endpoint, so the plugin estimates tokens from character length (chars ÷ 4). The token-usage indicator is approximate.
• Daemon down? — If the daemon stops, agent calls will surface a "Could not connect to the Ollama daemon" notice. Restart with ollama serve and click Refresh model list.