# Configuration

config.toml keys and environment variables.

Configuration is optional. Create `~/.config/bookmark-context/config.toml` to
override the defaults:

```toml
[daemon]
port = 7331

[embedder]
model = "BAAI/bge-small-en-v1.5"

[trace]
processes = false
mcp_tools = false
```

| Key | Default | Env override |
|---|---|---|
| `[daemon] port` | `7331` | — |
| `[embedder] model` | `BAAI/bge-small-en-v1.5` | — |
| `[trace] processes` | `false` | `BOOKMARK_CONTEXT_TRACE=1` |
| `[trace] mcp_tools` | `false` | `BOOKMARK_CONTEXT_MCP_TRACE=1`, or `BOOKMARK_CONTEXT_TRACE=1` |

Environment variables win over the config file. Changing the embedder model
means re-indexing every bookmark — existing vectors were written by the old
model.

## Choosing an embedder

The default `BAAI/bge-small-en-v1.5` is a small English model that runs locally
on CPU via [fastembed](https://github.com/qdrant/fastembed) — no API, no key. Any
model fastembed supports works here.

| Model | Vector dim | Notes |
|---|---|---|
| `BAAI/bge-small-en-v1.5` | 384 | default — fast, low disk |
| `BAAI/bge-base-en-v1.5` | 768 | better retrieval, ~3× slower to index |
| `sentence-transformers/all-MiniLM-L6-v2` | 384 | non-BAAI alternative, comparable size |
| `intfloat/e5-small-v2` | 384 | non-BAAI alternative |

BAAI (Beijing Academy of Artificial Intelligence) was added to the US Commerce
Department's Entity List in March 2025. That restricts US exports *to* BAAI; it
does not affect downloading or running their open-weight models, which execute
fully offline here. If your organization's policy nonetheless bars models from
Entity-Listed or state-affiliated developers, switch to one of the alternatives
above.

See [Tracing & debugging](/docs/guides/tracing/) for what the trace options record.