AI modes
Three ways to run AI in Geneziz - local via Ollama, the hosted Geneziz Cloud proxy metered in credits, or bring your own Anthropic or OpenAI-compatible key.
Three ways to run AI
Geneziz uses AI to enrich items (category, tags, description, priority) and to answer questions over your library. You choose where that AI runs:
| Mode | What it uses | Key needed | Credits |
|---|---|---|---|
| Local | Ollama on your machine (http://127.0.0.1:11434) | None | No |
| Hosted | Geneziz Cloud proxy | None | Yes |
| BYOK | Anthropic, or any OpenAI-compatible endpoint | Provider key | No |
Pick a mode in Settings - AI provider during first-run setup or later.
Local (Ollama)
Point Geneziz at a local Ollama instance (default http://127.0.0.1:11434). Enrichment runs on your machine, no API key is involved, and nothing leaves it. Local mode is available after purchase.
Hosted (Geneziz Cloud)
The proxy mode needs no key: Geneziz Cloud authenticates with your license key and device token and meters usage in credits. Each AI-enriched item costs roughly one credit; asking a question over your knowledge base (ask, and ask_kb through MCP) costs 1 credit. The $99 license includes 2,000 credits; top-up packs are listed on pricing and refunds.
Bring your own key (BYOK)
Bring an Anthropic key or an OpenAI-compatible endpoint (OpenAI, OpenRouter, GLM, or any compatible server). Calls go straight from your machine to the provider - unlimited processing, no credit consumption, no Geneziz Cloud in the middle.
Search is separate from generation
Search runs locally in every mode:
- Full-text keyword search - SQLite FTS5 with BM25 ranking. Available from the CLI without a license.
- Semantic search - local ONNX MiniLM embeddings, finds items by meaning. Requires an activated license.
geneziz search "react performance" # keyword
geneziz search "rust async runtime" --semantic # meaning-based
See the vault for where the index lives, and /landing for the product details.