Krate Company Brain Memory Query overview

Workflow for querying and updating the company brain knowledge base within Krate. Query flow: 1. An agent session (or human user via kubectl) creates an AgentMemoryQuery resource specifying search terms, filters (time range, source repository, author), and similarity threshold. 2. The aggregated API server resolves the query against the vector index in the AgentMemoryRepository, performing semantic similarity search combined with metadata filtering. 3. Matching AgentMemorySnapshot resources are returned, ranked by relevance. Each snapshot contains structured knowledge — code patterns, architecture decisions, error resolutions. 4. The agent incorporates relevant snapshots into its context before proceeding with its task. Update flow: 1. On agent session completion, the dispatch controller generates a new AgentMemorySnapshot summarizing decisions made, patterns discovered, and lessons learned. 2. The snapshot is embedded (vector encoding) and stored in the AgentMemoryRepository. 3. Org-admins can review, curate, or delete snapshots via the memory management UI or kubectl.