Measure AI consensus programmatically

When independently trained models agree, the answer is safer. When they split, check it before you act. Synero's API returns that signal as a structured object with a 0-100 score.

Get an API key View API docs

Free account: 3 full queries, no card. Then pay as you go from $5 of credit, no subscription.

POST/api/query

Returns individual model responses, a synthesis, and a consensus object (score, unanimous points, contested points, dissenter) in JSON mode or as an SSE event.

Your app

POST /api/query

Synero

Asks in parallel

GPTClaudeGeminiGrok

Combined answer

Plus where they agree

Capabilities

01Consensus Measurement

Models answer the same question independently. The consensus object lists the points all of them agreed on and each contested point with every model's stance.

02Decision Automation

Build automated decision trees based on consensus level. High consensus: auto-approve. Partial consensus: route to a specialist. No consensus: escalate to a human.

03Structured Output

The consensus object has typed fields: score, unanimous, contested, dissenter, and takeaway. Build logic on fields, not raw text.

04Domain Flexibility

Measure consensus on any question type: factual claims, strategic assessments, technical evaluations, risk analyses, or creative outputs.

05Your Own Tracking

Synero does not track consensus over time for you, but each response carries a queryId and score you can log and chart yourself.

06Agent Integration

Use the MCP server or the OpenClaw terminal agent skill to run consensus checks inside agentic workflows before high-stakes decisions.

Code Example

// Use consensus to automate decision quality gates
const result = await fetch('https://synero.ai/api/query', {
  method: 'POST',
  headers: {
    'Authorization': 'Bearer YOUR_API_KEY',
    'Content-Type': 'application/json',
  },
  body: JSON.stringify({
    format: 'json',
    preset: 'council',
    prompt: `Is this claim accurate? "${claim}"

Rate your confidence and identify any caveats or conditions.`,
  }),
});

const { consensus } = await result.json();
// consensus is null when fewer than two models answered or grading failed: treat that as unverified.
// consensus.score is 0-100; consensus.contested lists the splits
// - High score: auto-approve
// - Middle: approve with caveats noted
// - Low: flag consensus.contested for human review

Frequently asked questions

What does 'consensus' mean in this context?

Consensus means independently trained models from different providers reach the same conclusion on the same question. The API reports it as a 0-100 agreement score with the shared and contested points.

Can I tune the consensus threshold?

Synero returns the full responses, the synthesis, and the consensus object. You choose the score threshold that counts as sufficient for your application.

Is this useful for agentic workflows?

Yes. An agent can proceed when the score is high and escalate to a human when the models split. See the MCP server and agent hub for setup.

How does this compare to ensemble methods?

Traditional ensembles average or vote. Synero writes one synthesized answer and separately reports where the models split and what each model said.

Get an API key for LLM consensus

Four models. A structured agreement score. Free to try, no card.

Get an API key