Try it like production sees it.

Patch a model and a strategy, press Run, and read the lineage. The demo below runs the same pipeline shape as production — the real playground lives in your dashboard after install.

Playground demo

Idle

Model

Strategy

Routed response

Pick a model and a strategy, then press Run.

Illustrative demo — the real playground lives in the dashboard. Docs →

Four moves to a working prompt.

  1. Patch model + strategy

    model: gpt-4o-mini · strategy: priority

    Choose from your registered models and the five routing strategies.

    Model registry

  2. Send the prompt

    max_tokens: 256 · temperature: 0.3

    Try the exact prompt your application will send.

    Playground docs

  3. Read the lineage

    synapass: { provider, fallback_used, latency_ms, estimated_cost_usd }

    Every trial returns who answered, whether fallback fired, latency, and cost.

    Lineage

  4. Promote to an endpoint

    endpoint: support-draft · policy: eu-support

    Bind the prompt to a named endpoint with limits — no model names in app code.

    Endpoints

Full walkthrough: Playground docs

The real playground is one install away.

One command starts the full stack. Your first routed call lands in about five minutes — with the lineage to prove it.

Get startedStar on GitHub