Products · Better RAG

View Demo: Better RAG

Test a Better RAG is a live UI against a running Amnesis API: real ingest, real recall, and chat that shows when answers are memory-backed (Sources) vs when the system is answering without memory—plus the model and policy used for the turn.

What you’ll do

1. Open Cockpit — After sign-in, go to /app. Create an Avatar if needed, then open it: Ingest (corpus), Memory (recall + node inspect), Test (Guide / User Avatar chat with Sources), Deploy / Integrate (endpoint + session token for your app or site).
2. Sign in — Header Sign in opens the drawer. Register or log in. If your token has no workspace, pick one from the list or paste a workspace UUID.
3. Ingest — On Ingest, upload a small file (PDF or text). Wait for the API response; expand details if you want raw JSON.
4. Memory — On the Memory tab, run semantic recall and load nodes; use Inspect for JSON. The right Tools rail lists ingested files, revisions, and diff — remove or re-ingest from there when you change a source document.
5. Test — On the Test tab, run Guide vs User Avatar chat. Check Sources this turn in the right rail and the role chip on each answer.

What “Better RAG” proves here

• Scoped memory

  Everything runs in your workspace—not a generic index you can’t audit.

• Inspectable retrieval

  Recall returns structured hits you can open; chat shows how much context was injected.

• No fake success

  Errors and empty hits come straight from the API—there is no demo-only “always green” path.

Running locally

First (recommended): run the stack from the repository (e.g. ./scripts/run.sh), sign in, then open /app on the same origin and follow the steps above in that UI. Better RAG evaluations should work through the UI first.

Second (optional): Dev API console for raw fetch calls from the browser. Shell curl, seed scripts, and other CLIs are for operators and future CI / automation—not the default path for evaluators.

Open Better RAG Why Better RAG How it works