Total Recall Smart Agents

Closed tickets become
the next right escalation.

A support bot learns recurring issue categories automatically. L1 gets a practical routing hint. Admins can inspect storage proof and veto a rule.

24 fictional closed ticketsAuto-learn: ≥2 tickets · ≥60% agreementWalrus Memory + Tatum

Sunrise Beverages · Fictional systems and tickets
Your teaching and admin vetoes stay in your demo session.

01 / SHARED EXPERIENCE

What the bot learned

Rules emerge from closed tickets. No approval queue.

Loading learned rules…

Teach from a closed ticket

Use a fictional closed ticket. A single new issue stays as evidence until the confidence threshold is met.

02 / EVERYDAY SUPPORT

L1 asks

L1 sees only this.

Try the same question with memory OFF and ON to see how the routing changes.

03 / VERIFY & VETO

Admin proof

L1 never sees this.

Storage healthWaiting for an epoch check…

Proofs appear with the learned rules.

“Block rule” is the demo equivalent of an admin #sai veto. “Reopen rule” mirrors #mo. Both affect only your session and persist through Walrus Memory.

Does memory improve routing?

10 new fictional tickets · same model and instructions · one batch per mode

Runs two LLM calls, with one retry if needed. Scoring labels stay on the server. One run per minute per IP.

Total Recall · Tatum × Walrus Memory · Synthetic demo data only. Storage indexing and proof may take a little time. Measured results depend on the configured model; no scores are prefilled.