Skip to main content
Kioku is built around three core concepts: knowledge, sessions, and meetings.

​
Knowledge

Knowledge is the searchable corpus that Kioku builds from your data. It lives in two forms:
  • Documents — PDFs you upload. Text is extracted, chunked, embedded via Ollama, and stored in Qdrant.
  • Meetings — transcripts from recorded meetings. Each transcript segment is embedded and indexed.
All knowledge is searchable via vector similarity: POST /knowledge/search with a query returns the most relevant chunks ranked by semantic score.
Kioku uses nomic-embed-text-v2-moe for embeddings — a model that matches OpenAI’s text-embedding-3-small on benchmarks while running locally on your hardware.

​
Sessions

Sessions are conversation containers. Each session has:
  • Messages — user and assistant messages in OpenAI chat format (multi-part content)
  • Traces — execution records that track what happened during message processing
  • Mode — the session’s purpose (e.g. “research”, “chat”)
Sessions are the primary way to interact with Kioku’s API programmatically.

​
Meetings

Meetings are ingested from Vexa’s bot platform. The lifecycle:
  1. A bot is requested via POST /vexa/bots
  2. The bot joins the meeting (Google Meet, Zoom, or MS Teams)
  3. Audio is captured and transcribed in real-time via Whisper
  4. When the meeting ends, the transcript is sent to Hivemind
  5. Transcript segments are embedded and become searchable knowledge

​
MCP

The Model Context Protocol (MCP) server exposes Kioku’s knowledge as tools that AI clients can call:
  • kioku_search — search the knowledge base
  • kioku_list_meetings — list all meetings
  • kioku_get_transcript — get a meeting’s transcript
  • kioku_list_documents — list uploaded documents
This lets Claude, Cursor, and other MCP-compatible clients directly access your meeting context.

​
Companies and Members

Kioku supports multi-tenant organization:
  • Admin — creates the company, manages members and API keys
  • Members — invited by admin, can use sessions/knowledge/search
  • Personal — standalone accounts without a company
API keys can be scoped per-company and per-user. CLI auth keys are long-lived tokens for terminal access.