3-Minute Quickstart

Try ACP instantly with no installation. See how multiple AI models converge on a single answer through axiom-grounded consensus.

Playground status

The web playground is currently under development. For immediate hands-on access, use the API directly (see the API Quickstart) or run the Dashboard locally via the Full Setup guide.

Option A: Demo Mode

Demo Mode lets you explore a pre-recorded consensus run without any API keys or account setup. It is the fastest way to understand what ACP does.

Visit axiomprotocol.org/playground
Select the Demo Mode tab
Click any example query or enter your own
Click Run Consensus (Demo)
View the pre-recorded consensus result

What you will see

The demo visualizes the full consensus lifecycle. Here are the key elements on screen:

Element	Description
D-score	Divergence between models. Drops from ~0.35 to ~0.08, indicating consensus was reached.
Final Answer	The synthesized answer that all models agreed upon after iterative convergence.
Axiom Grounding	Shows which of the 7 axiom levels were referenced to anchor the consensus.
Iteration Timeline	A step-by-step view of how model responses converged across iterations.

Key insight

Watch the D-score timeline closely. ACP uses φ-convergence -- each axiom level reduces divergence by the golden ratio, so consensus is mathematically guaranteed to converge.

Option B: Live Mode via API

If you have an OpenRouter API key, you can run a live consensus request right from your terminal. This sends your query to multiple models simultaneously and returns the converged result.

Run a live consensus request

curl -X POST https://your-worker.workers.dev/consensus-iterative \
  -H "Content-Type: application/json" \
  -H "x-openrouter-key: YOUR_KEY" \
  -d '{
    "query": "What is 2+2?",
    "models": ["openai/gpt-5.4-mini", "anthropic/claude-haiku-4-5"]
  }'

Replace YOUR_KEY with your OpenRouter API key. You can sign up for one at openrouter.ai/keys.

Cost estimate

Each consensus query costs approximately $0.05 -- $0.15 depending on the models you select and the number of iterations needed. Simple factual queries with fast models (like GPT-4o-mini and Claude 3 Haiku) are at the low end; complex reasoning queries with larger models cost more.

Understanding the result

Whether you use Demo Mode or Live Mode, the output contains the same structure. The most important fields are:

Field	Type	Meaning
`final_D`	number	Divergence score. 0 = perfect consensus, 1 = total disagreement.
`consensus_reached`	boolean	True when D-score falls below the threshold (default: 0.1).
`final_answer`	string	The consensus answer synthesized from all model responses.
`iterations_used`	number	How many rounds of convergence were needed.

Next steps

API Quickstart (10 min) -- integrate ACP into your own application with code examples for fact-checking, code review, and more.
Full Setup (30 min) -- run the complete ACP stack locally for development and exploration.
φ-Convergence -- understand the mathematical foundation behind D-score reduction.

Getting Started

API Quickstart