Bay Run / playground
Minting a zero-secret demo token… Free tier · waiting for live rate-limit headers

Live service statistics

Free tier
60 req/min · no signup
Registered
120 req/min free
Conformance
loading…

Measured on this instance since start, refreshed from /status.json.

Zero signup · real API · 10 seconds

Feel the route.

Give Bay Run one task. Watch it choose a specialist, stream a small model, and sign the result. The controls below call the same free API you can ship against.

Click Run once in each panel. Everything is prefilled; edit the samples when you want to make the proof yours.

01 / real request

Route a task

Pick a normal task class. Bay Run chooses the model and explains the decision in the response.

Copy the current request without exposing the demo token.

Waiting for the demo token…

02 / SSE stream

Chat (streamed)

Open the stream and watch each token arrive from the Qwen3 GPU lane. Chat runs only when that lane is warm or warming; otherwise the API returns structured no_specialists.

Copy the current streamed chat request.

Waiting for the demo token…

Your streamed answer will appear here.

Model: waitingTokens: —Latency: —Free tier · no card, no signup

SSE: idle

03 / prove it on examples

Mini bake-off

Three labeled examples are enough to compare curated candidates. The response publishes a permanent public result and signs the scorecard.

Copy the current labeled examples.

Waiting for the demo token…

04 / state across calls

Memory

Your agent remembers across calls. The same zero-signup token below writes a key and reads it back on a separate request — private to that token, and expiring with it.

Copy the current memory request after choosing Remember or Recall.

Waiting for the demo token…

Session memory is capped and ephemeral by construction. Registering a client or minting a bay_live_ key is the durable upgrade — same API, longer retention and a larger row budget.