DruxAI
REST API · No SDK required · Free to try

Stop trusting
one AI.

The Drux Consensus API fans your prompt out to multiple frontier models simultaneously, scores how much they agree, and returns a single synthesised answer with a confidence score. One call. Real signal.

3 models · simultaneously

Llama 4 MaverickMeta
Agrees

Remote work generally improves productivity for individual contributors. Studies show 13-20% output gains when distractions are managed. The key variables are home setup quality and task type.

DeepSeek V3DeepSeek
Agrees

The productivity question depends heavily on role. Engineers and writers thrive remotely. Sales, mentorship, and creative brainstorming benefit from physical presence. A hybrid model captures both.

Gemma 3 27BGoogle
Diverges

Evidence is genuinely mixed. Remote work increases autonomy and reduces commute stress, which correlates with output. But collaboration overhead rises and junior employees develop skills more slowly without osmotic learning.

Consensus output

72

Leans one way

Consensus score / 100

ContestedSplitConverging

Synthesis

Most models agree remote work can boost individual focus and deep work, but the evidence is mixed for collaborative tasks. The consensus leans positive for knowledge workers with structured routines, while flagging that team cohesion and onboarding suffer without intentional in-person time.

Preview — click Run to try your own query

How it works

01

One API call

Send your prompt once. We fan it out to all requested models in parallel — no serial waits, no rate-limit juggling.

02

Parallel execution

All models run simultaneously. Results come back in seconds, not minutes. We handle retries, timeouts, and fallbacks.

03

Consensus score

We embed all responses, cluster by semantic similarity, and compute Shannon entropy. The result: a 0–100 confidence score and a single synthesised answer.

Integrate in minutes

curl -X POST https://drux.space/api/v1/consensus \
  -H "Authorization: Bearer drx_your_key" \
  -H "Content-Type: application/json" \
  -d '{
    "query": "Is remote work better for productivity?",
    "models": [
      "meta-llama/llama-4-maverick",
      "deepseek/deepseek-chat",
      "google/gemma-3-27b-it"
    ]
  }'

Built for real use cases

🔬

Research & fact-checking

Cross-verify claims across multiple frontier models. Low consensus flags contested or uncertain information before you publish.

🏗️

Product decisions

Run technical or strategic questions through multiple models and score how much they agree. High consensus = strong signal.

📊

Content moderation

Multi-model scoring dramatically reduces false positives. A single model's blind spots get corrected by the group.

🤖

Agent pipelines

Use consensus as a confidence gate. Only proceed when models agree above a threshold. Reduce hallucinations in production.

💬

Customer support

Route ambiguous tickets to human review when AI consensus is low. Automate only the high-confidence answers.

📝

Document review

Surface sections where models disagree on interpretation — contracts, policies, medical texts. Disagreement = review flag.

Reading the score

0 – 25

Highly contested

Models fundamentally disagree. Treat this topic as unresolved. Human review recommended.

26 – 50

Divided

Meaningful disagreement. Multiple valid perspectives exist. Useful to surface the splits.

51 – 75

Leans one way

General agreement with notable exceptions. The synthesis is reliable but worth reading.

76 – 100

Converging

Strong consensus across models. High confidence in the synthesised answer.

Simple credit pricing

1 credit per model

A 3-model consensus call costs 3 credits. A 5-model call costs 5. No subscriptions, no minimums. Credits never expire.

400

credits for $5

1,000

credits for $10

3,000

credits for $30

Start free — no card required →

Ready to build with consensus?

Get your API key in 30 seconds. Free tier included.