Evidence in. A defensible decision out.
Setliva is one queue for chargebacks and overdue invoices, and an AI agent working it. Open any case and read exactly why it decided what it did.
RCV-8821 | 141 days past due | To a human |
RCV-8803 | 74 days past due | To a human |
RCV-8819 | 67 days past due | To a human |
RCV-8825 | 58 days past due | Payment plan |
RCV-8807 | 38 days past due | Email reminder |
RCV-8823 | 33 days past due | WhatsApp nudge |
DSP-4402 | Window closed | Accept liability |
DSP-4441 | 33 hours left | To a human |
Five stages, each one you can inspect.
It reads 25 kinds of record and can draw 52 distinct facts from them. The models read and draft, but never decide alone — and the same case always produces the same digest.
Extract
Reads the systems of record and derives facts.
Every fact carries the record it came from. If a record is absent the fact is absent — and that silence is scored, not ignored.
Weigh
Scores facts against your thresholds, not ours.
Corroborating evidence of the same kind discounts rather than stacks: three proofs the buyer was real are not three times one.
Decide
Picks the next action inside your policy.
15 hard gates bypass the score entirely, so a disqualifying fact can never be averaged away.
Produce
Writes the representment or the message.
Every sentence is bound to a cited fact. There is no free composition, so it cannot assert what the evidence does not contain.
Self-check
Verifies before anything is released.
Every citation must resolve to a fact, every fact to a record, and every amount to one or the other. Otherwise it is withheld.
Every weight, every source, on one screen.
Approve, override or return a decision with a note. Your verdict joins the audit chain next to the engine’s own reasoning, and the next case is one keystroke away.
- Facts grouped for and against, each traced to a record
- The gate that decided the case, when one did
- The generated artifact with its exhibit list
- Keyboard-first: J and K for next and previous, A to approve
- +4AVS full matchEV-4431-04
- +3.2CVV2 matchEV-4431-04
- +30Authenticated with liability shiftEV-4431-05
- +16Product used after purchaseEV-4431-02
- +4Licences in active useEV-4431-02
Everything a recovery team works from.
Working queue
Everything open, sorted by time left to act. Filter, search and save the views your team works from.
Case workspace
The records, the facts and their weights, the gate that fired, the artifact, and a one-click decision.
Quality review
Where the engine is least sure: cases near the floor, self-check findings, and human overrides.
Analytics
Outcomes, gates, score distribution and exposure by currency — computed live, never cached.
Audit trail
A hash-chained log of every decision and every human act, verified on load.
Policy
The thresholds the agent decides against, on one page, owned by the person accountable.
The systems your evidence already lives in.
The engine reads records, not vendors, so adding a source never changes a decision rule. Ingestion for your stack is scoped with you during onboarding.
The demo book carries records in each of these systems’ own formats. Named descriptively — not an endorsement or partnership.
Recover what you are already writing off.
Open the demo workspace and work a real queue end to end — the evidence, the decision, the representment it produces. No card, no call.