Local retrieval
This release uses lexical scoring in your browser, not an external model. It never claims semantic AI accuracy.
GROUNDED ASSISTANTS
Inspect the knowledge,
not just the answer.
Build a cited support assistant from text you control. Every answer shows the exact source chunks it used.
PLAYGROUND
Ask, inspect, correct
Optional BYOK model pass
This release prepares requests but does not send provider calls or spend money.
Ask a question to see a deterministic local retrieval trace.
QUALITY REVIEW
Turn conversations into test cases
Evaluation suite
Conversation review
ANALYTICS
See the gaps behind the score
Question themes
Source coverage
Failure queue
Safety and handoff
TRUST
The boundary is visible
Injection warnings
Source text is scanned for instruction-like phrases and flagged for review, not silently obeyed.
Portable data
Export all sources, chunks, conversations and evaluations. Delete them in one action.