Context Engine Design

The Context Engine extracts knowledge from every query response and uses it to enrich future queries.

What It Does

1. Extract — After each response, the Context Engine identifies key facts (atomic facts) in the response
2. Embed — Each fact is embedded using a local embedding model and stored in the knowledge base
3. Retrieve — When a new query arrives, the Context Engine searches for relevant facts and enriches the query with them

How It Enriches Queries

When a new query arrives:

1. The query is embedded
2. The Context Engine searches for similar facts in the knowledge base
3. Relevant facts are added to the query context
4. The enriched query is sent to the LLM

This means the LLM receives not just the user's query, but also relevant context from previous interactions — without the user having to provide it.

Document Ingestion

In addition to extracting facts from responses, the Context Engine can ingest external documents:

1. Navigate to Context in the sidebar.
2. Upload documents (text, markdown, PDF).
3. The Context Engine extracts facts from the documents.
4. These facts are available for enriching future queries.

Privacy and Data Isolation

• All facts are scoped to the organization — no cross-organization data sharing
• Facts are stored in the organization's database
• Embedding generation runs locally — no external API calls