Grounded chat

Answers with receipts.

One surface, three audiences. Engineers, PMs, and clients each get a role-scoped answer with inline citations back to the exact source chunk.

↓ scroll

Three live demos · scroll through each

Same project. Three audiences. Three full-screen views.

Click into any iframe and ask a question. Watch the SSE stream and the citation chips resolve in real time. Each demo runs in its own viewport — scroll to walk through them in order.

Live demo unavailable

Run npm run db:migrate && npm run db:seed to mint a demo share token, then refresh this page.

Live demo unavailable

Run npm run db:migrate && npm run db:seed to mint a demo share token, then refresh this page.

app.example.com/demo
DEMO · KYC Portal

Live demo unavailable

Run npm run db:migrate && npm run db:seed to mint a demo share token, then refresh this page.

Heads up

The demos run against a seeded sample project. Tokens rotate every 7 days, scoped to client audience. No data leaves your browser; the SSE stream proxies through the same Postgres-native context engine your production workspace uses.

Capabilities

What the chat surface ships with.

Inline citations

[1] src/auth.ts:42, [2] sprint-plan.pdf §3. Every chip is clickable; admin routes to the source, public surface shows a preview.

Audience-scoped retrieval

Phase 18's role-aware retriever gates code, memory, and similar features by role. Clients never see internal-only sources — even via prompt injection.

Token-accurate billing

Input + output tokens persisted per turn, rolled up nightly into chat_input / chat_output line items. No silent over-runs.

Streaming-safe citation parser

Tail-buffered ≥16 chars so [1] markers split across SSE deltas still resolve. Works under burst load.

Start chatting with your codebase.

Connect a repo, ask any question, watch every claim cite its source.