Setup

Open Settings > Agentic Chat after you install the plugin, then pick a provider on the Models tab.

OpenRouter

OpenRouter is the default provider. The default model is moonshotai/kimi-k2.6.

1. Create an API key at openrouter.ai/keys.
2. Paste it into Settings > Agentic Chat on the Models tab.
3. Pick a model that supports tools. Model browsing shows only options that satisfy the zero-data-retention routing requirements while the strict defaults are on.
4. Keep the strict privacy routing defaults unless you want a broader provider set. Strict routing can mean a model has no compliant endpoint. In that case, pick another model, relax Require zero data retention or Deny prompt logging (prompts may then be logged or used for training), or use Ollama.

First chat

1. Click the ribbon icon or run Agentic Chat: Open chat.
2. Open a note so it attaches automatically, then send a prompt such as Summarize @Meeting-notes.
3. The inline step card shows which notes the agent read before it answered.
4. Ask for a small edit, approve the diff, then run /undo to revert it.

If the model list is empty, see Troubleshoot.

Ollama

Use Ollama when you want fully local model calls.

1. Run a local Ollama server.
2. Pull a model that supports tools.
3. Set the provider to Ollama (local).
4. Keep the default server URL, http://localhost:11434, unless your Ollama server uses another address.
5. Enter the Ollama model tag.

Ollama requests stay on your device unless your own Ollama setup routes them elsewhere.

OpenAI-compatible gateways

Choose OpenAI-compatible for gateways such as OpenWebUI, LM Studio, vLLM, llama.cpp, Chutes, or Venice.ai.

Set the base URL to the gateway root that serves /chat/completions. For OpenWebUI you can use the site root or the API base directly, because the plugin resolves a bare OpenWebUI root to /api. The default local OpenWebUI API base is:

http://localhost:3000/api

Then paste the gateway bearer token and the model id that the gateway exposes.

Thinking level

Some models support adjustable reasoning effort. The default is off. Set it in Settings > Agentic Chat on the Agent tab, or per message with /effort [level] in chat:

Level	When to use it
off	Default. Simple questions and small edits.
minimal, low	A small amount of reasoning for routine tasks.
medium	Multi-step vault tasks.
high, xhigh	Hard planning and research. These levels use more tokens.

Other models ignore the setting.

Proxy settings

On desktop, Models > Network proxy > HTTP proxy lets plugin-owned model, model-browsing, web, MCP, and observability requests use an HTTP proxy. Enter http://host:port with no trailing slash. The plugin accepts only the http scheme.

On mobile, leave the plugin proxy fields empty and use the device, VPN, or network proxy.

Semantic retrieval setup

If you want vector-based note search, enable Settings > Resources > Semantic retrieval:

1. Choose an embedding provider (OpenRouter, Ollama, or OpenAI-compatible).
2. Enter the embedding model id.
3. Set vector dimensions to match the model (default 1536).
4. Use Ollama for fully local embeddings if you do not want note content sent to a remote provider.

After setup, run /semantic-index estimate to preview note and token counts, then /semantic-index start to build the index for your current scope.