A scholarly instrument for AI deliberation

Convene a council
of models.

WitanWeave doesn't ask one model and hope. It convenes several — they generate, verify, cross-examine, and adjudicate under a deterministic state machine. Every claim, every ruling, on the record.

Enter the council → Invite-only during early access.
The deliberation

A verdict you can audit.

A deterministic orchestrator — never an LLM — runs the state machine. Models supply content; the harness supplies the process, and records every step.

01

Frame

The question is scoped and budgeted before a single model is called.

02

Generate

Multiple models answer independently — no groupthink, no lead voice.

03

Extract & verify

Claims are pulled out and checked against evidence, in a sandbox.

04

Cross-examine

Models critique each other; clashes escalate, agreement settles cheaply.

05

Adjudicate

Evidence-first rulings compose the final answer — with the receipts.

Why it's different

Rigor, on your terms.

Bring your own key

Point it at Claude, GPT, Grok, Gemini. Local models run free for everyone. Your keys, your spend, never logged.

Replayable & auditable

An append-only event log is the system of record. Every run replays byte-for-byte — the whole deliberation is evidence.

Deterministic by design

The orchestrator is a state machine, not a model. Control flow never bends to a prompt; budgets and circuit-breakers hold.

“The counsel of many, weighed by one steady hand.”

— the design principle of the Witan