Moamao documentation
Moamao is a mixture-of-agents API. You send one prompt; a panel of up to four models from different providers (claude, gpt, gemini, and llama families) drafts answers in parallel, each model reads and critiques the others' drafts, and a judge model synthesizes one final answer — with a full trace of who said what.
Open beta. The API is live and surfaces may change. Breaking changes are announced in the changelog. To get started, create a free account and generate a key in the dashboard.
When to use a mixture
A mixture run costs more and takes longer than a single model call — roughly the sum of the member calls plus the judge, in about the wall time of two sequential calls, since each layer runs in parallel. Use it when the answer matters more than the latency:
- High-stakes output— migration plans, contract summaries, code review, anything where one model's hallucination is expensive. Cross-critique means a weak claim rarely survives three other models' review.
- Open-ended or contested questions — research, analysis, and long-form writing, where models trained on different data genuinely disagree and the disagreement is signal.
- Auditable answers — every run keeps a replayable trace, so you can show exactly which model contributed what.
When a single model is enough
Lookups, formatting, classification, short transformations — anything a fast model gets right on the first try. You don't have to decide per request: smart routing detects easy prompts and sends them to one fast model, billed as a quarter of a run. See Concepts for how routing decides.
Map of the docs
- Quickstart — get an API key, add your provider keys, and make your first run with curl or the TypeScript SDK.
- Concepts — the vocabulary: runs, panels, proposers, layers, aggregators, traces, smart routing, and presets.
- API reference — endpoints, request and response fields, error codes, and beta rate limits.
Questions the docs don't answer? Write to hello@moamao.com.