Curate and approve knowledge entries that enhance AI responses with agency-specific context and domain expertise.
Upload knowledge entries individually or in bulk via CSV/JSONL. Each entry has a title, content, category, and source reference. The upload process validates format and content length, then queues entries for Vectorize embedding so they become searchable by the AI system immediately.
Knowledge entries go through an approval workflow before they influence AI responses. New entries start as pending, reviewers can approve or reject with comments, and only approved entries are included in the AI context retrieval pipeline. This prevents inaccurate or outdated information from degrading AI quality.
When new entries are uploaded, the system checks for semantic duplicates using Vectorize cosine similarity. Entries that are too similar to existing ones are flagged for review rather than auto-approved. This prevents the knowledge base from accumulating redundant information that wastes context window space.
Six API endpoints handle the full knowledge lifecycle — list with search and filters, create single entries, bulk create from file, read individual entries with full content, update entries and re-embed, and delete entries with Vectorize cleanup. The admin UI provides a searchable, filterable interface for managing hundreds of knowledge entries.