Configure vector search

Vector search needs an embedding model, a stable profile ID, and the model's output dimension.

1. Set one embedding profile

export POWERCONTEXT_SERVER_INFERENCE_EMBEDDING_MODEL=provider:embedding-model
export POWERCONTEXT_SERVER_INFERENCE_EMBEDDING_PROFILE_ID=embedding-model-v1
export POWERCONTEXT_SERVER_INFERENCE_EMBEDDING_DIMENSION=1024

Replace the example values with one supported provider model, a profile ID you keep stable for that model, and its documented output dimension. POWERCONTEXT_SERVER_INFERENCE_EMBEDDING_NORMALIZATION defaults to unit.

The output dimension is always required. It fixes the local vector index width and validates returned vectors. By default that value is also sent as the dimensions request field. Some OpenAI-compatible services reject the field, including SiliconFlow BAAI/bge-m3. Keep the documented output dimension (1024 for bge-m3) and disable sending:

export POWERCONTEXT_SERVER_INFERENCE_EMBEDDING_SEND_DIMENSIONS=false

2. Start the Server

powercontext server run

With SQLite, PowerContext loads the bundled sqlite-vec extension when it opens the database. Startup fails if the installed extension is incompatible with the platform or SQLite build.

3. Verify the capability

powercontext capabilities

The result reports the enabled search modes. Without an embedding profile, SQLite full-text search remains available.

The capability flag reports a loaded channel; inspect Source extraction, projection writes, and actual search hits separately. See Configure models. Search identities are in hits[].memory.artifact, and vector hits include vector in matched_by. Explicit vector/hybrid modes fail when vectors are unavailable rather than silently falling back.

When enabling or changing an embedding profile, stop normal service and rebuild the current projection using the Atomic Memory migration guide. Rebuilding preserves Artifact identities, content revisions, and state versions. Do not change a profile ID or dimension just to bypass a mismatch.

Atomic Memory currently uses exact L2 distance for ordinary vector search and extraction threshold enumeration. Search limits returned results; extraction returns every eligible threshold match. Work grows with eligible vector count and dimension, independently of the response limit.

For timeouts, batch size, storage settings, and exact defaults, see Configuration.