Multi-model deliberation

Pick a panel. Watch them deliberate.
Fuse the best answer.

Fusion runs several models side-by-side on your prompt, then a judge produces a structured analysis — consensus, contradictions, blind spots — and fuses one best answer. You see every step. The transparent, configurable counterpart to Adamant.

How it works

Three steps, nothing hidden

Fusion is designed to be auditable end-to-end. You pick the models; you see the deliberation; you get the fused answer with the full trace.

Pick a panel

Choose which models run on your prompt. Fusion dispatches them in parallel — no sequential bottleneck, no hidden routing.

Watch them deliberate

A judge model reads every response and produces a structured analysis: areas of consensus, contradictions, and blind spots. Nothing is hidden.

Receive the best answer

The judge synthesises one best answer from the panel. You get the final response plus the full deliberation trace — every step shown, every model credited.

Fusion vs Adamant

Transparent and configurable — by design

Adamant enforces governance policies over a model's own answers. Fusion is complementary: it puts multiple models in deliberation and surfaces the full reasoning chain so you can inspect exactly how the answer was formed.

Fusion

  • You choose which models join the panel
  • Full deliberation trace returned — every model's response visible
  • Best for research, drafting, and decisions where you want multiple perspectives

Adamant

  • Attach a declarative policy; Adamant enforces it deterministically
  • Every response includes the exact rules that fired and why
  • Best for compliance, content moderation, and accountable decisions
API

Two ways to call Fusion

Both surfaces are OpenAI-compatible. Switch from single-model to multi-model deliberation with a one-line change.

Model alias
modelgates/fusion

Drop into any OpenAI-compatible client. Pass your messages as usual; Fusion handles the panel behind the scenes.

Tool call
modelgates:fusion

Invoke Fusion as an agentic tool so your orchestrator can decide when to engage the panel and inspect the deliberation output.

See the deliberation yourself.

Open the playground, pick your panel, and send a prompt.