AI Features¶
Stratum integrates with LLM providers for AI-powered note operations, chat, and research.
Setting Up AI¶
Prerequisites¶
You need a running LLM provider. Options:
| Provider | Setup | Notes |
|---|---|---|
| Ollama (recommended) | Install Ollama, pull a model | Fully offline, free |
| OpenAI | Get an API key | Cloud-based, paid |
| Anthropic | Get an API key | Cloud-based, paid |
| Google AI | Get an API key | Cloud-based |
| Custom | Any OpenAI-compatible endpoint | Self-hosted or third-party |
Configuration¶
- Open Settings → AI
- Select your Provider
- Enter the API Endpoint (for Ollama/custom) or API Key (for cloud)
- Set the Default Chat Model
- Click Fetch Models to refresh the model list
- Assign capabilities to each model:
chat,embedding - Toggle RAG on for retrieval-augmented generation
- Click Save
AI Transform Actions¶
The AI can transform block content directly in the editor. Select text and choose an action:
| Action | Description |
|---|---|
| Rewrite | Improve clarity and flow while preserving meaning |
| Format | Clean up markdown, fix syntax, consistent headings |
| Structure | Organize notes into hierarchical sections |
| Summarize | Condense text while preserving key points |
| Connect | Add relevant [[wiki-links]] to related concepts |
| Generate Mermaid | Create a diagram from a text description |
AI Chat (Slash Menu)¶
Type / in the editor to open the AI slash menu. This gives you access to inline AI operations.
RAG Chat¶
When RAG is enabled, AI operations include context from your notes:
- Search your vault for relevant blocks
- Concatenate matching content as context
- Send the context + your prompt to the LLM
- Return results with citations to source notes
This means the AI answers based on your knowledge, not just its training data.
Interlink Notes¶
The Connect action scans a block and suggests [[wiki-links]] to related pages in your vault. This is useful for:
- Backfilling links when importing notes
- Discovering connections between separate topics
- Building out your knowledge graph automatically
Tips¶
- Ollama recommendation: Use
llama3.2for chat andnomic-embed-textfor embeddings - RAG chunk count: Start with 5 chunks. Increase for broader context, decrease for faster responses
- Model capabilities: Be intentional about which models get which capabilities — a small embedding model shouldn't be used for chat
- Privacy: With Ollama, everything runs locally — no data leaves your machine