Private knowledge that answers back.
Kanyarasi indexes Drive, Slack, and GitHub into one private, searchable corpus. Query it after the meeting or let it fact-check the room in real time — every answer cites its source.
Surfaces
Two ways to unlock your corpus.
Probe
After the meeting
Ask a question, get a grounded answer. Every response links to the exact Drive page, Slack message, or GitHub issue it came from. Built for the slower, deeper queries that resolve a decision after the fact.
- Inline citations on every answer
- Resumable conversation threads
- Source panel expands to the full chunk
Live
During the meeting
Start a session and Kanyarasi listens. When a claim lands, it checks your corpus and surfaces the evidence — or flags the gap — before the room moves on. Designed for the moment someone asserts a number.
- Real-time claim extraction and verification
- Per-turn evidence or explicit no-evidence signal
- Session replay with full turn history
Sources
Three sources. One grounded index.
Google Drive
Files and folders. Every chunk carries a stable page locator so answers link back to exactly where the fact lives.
Slack
Channels and threads. Speaker, timestamp, and channel travel with each chunk — so you know who said what and when.
GitHub
READMEs, issues, PRs, discussions, and releases — one connection indexes the full knowledge graph of a repository.
Retrieval
Built on evidence, not inference.
Stable citations
Every chunk preserves its source locator. Answers only cite what survived retrieval — no hallucinated references.
Hybrid retrieval
Dense pgvector + BM25 fused with Reciprocal Rank Fusion, capped per source so one noisy channel can't dominate.
Evidence-gated
The model cannot reference a source it did not retrieve. If the evidence isn't in the corpus, the answer says so.
Your knowledge stays private. Your answers stay honest.
Index your corpus once. Query it forever. No data leaves your control.