# Knowledge Graph & RAG

OX indexes your vault into a searchable knowledge graph with hybrid retrieval — BM25 full-text search combined with vector similarity — so both exact terms and semantic meaning rank together.

## Hybrid search

- **BM25 full-text** via Bleve for precise keyword matching.
- **Vector similarity** via Ollama embeddings (default model: `nomic-embed-text`) for semantic recall.
- Results from both are merged and ranked.

```
$ ox index
  Indexing vault at ~/ox/o/o/...
  Parsed 847 notes, 2,341 links, 892 tags
  Generated 847 embeddings (nomic-embed-text)
  Graph: 1,739 nodes, 3,233 edges
  Index time: 12.4s (incremental: 0.8s)

$ ox search "auth migration decision"
  0.94  o/when/days/2026.March.16.Monday.md
        "discussed auth migration with Alice..."
  0.87  o/what/orgs/COM/iwl/maxre/auth-decisions.md
        "JWT refresh token rotation strategy..."
  0.81  x/code/.../crm-api/CLAUDE.md
        "Auth middleware uses Supabase JWT..."
  Results: 3 (0.04s)
```

## The graph

A SQLite knowledge graph stores **nodes** (notes, tags) and **edges** (`LINKS_TO`, `HAS_TAG`). Wikilinks between notes and tags become traversable relationships, so you can ask not just "what matches" but "what's related."

## Pluggable backends

OX is built around clean interfaces — `GraphStore`, `VectorStore`, `TextSearchStore`, and `Embedder` — so the storage and retrieval stack is swappable.

- **Default stack:** SQLite + Bleve + Ollama (zero cloud dependencies, runs fully local).
- **Planned backends:** Neo4j, Qdrant, Pinecone, and OpenAI embeddings.

## Incremental & safe

- **Incremental indexing** with SHA-256 change detection — only re-index what changed (`ox index` re-runs in well under a second on an unchanged vault; use `--full` to force).
- **Read-only search mode** — no file-lock conflicts with your editor.
- **No embeddings needed?** `ox index --no-embed` skips Ollama and runs text-only.

## Next steps

- [Code Indexing](https://iamox.ai/docs/code-indexing.md) — index repos alongside notes
- [MCP Server](https://iamox.ai/docs/mcp-server.md) — expose search to AI agents
- [CLI Reference](https://iamox.ai/docs/cli.md#ox-index)