Dashboard playground; same gateway as the AI API, but authenticated with the session and billed to the user.
Playground chat completion
Same as POST /v1/chat/completions but authenticated
with the session and billed to the user directly (no key limits). Logged with source: playground.
Rate-limited at the per-key requests-per-minute limit. At most 20 inline images per request,
each up to 10 MB (encoded).
When streaming, after the upstream stream (including its data: [DONE]) one extra event reports
the billed usage:
data: {"object":"sumorouter.billing","usage":{"prompt_tokens":…,"completion_tokens":…,"cached_tokens":…,"cache_write_tokens":…,"cost":<USD>}}.
Playground chat completion › Request Body
modelModel slug from GET /v1/models (a model with mode chat).
streamStream server-sent events. stream_options.include_usage is forced on.
max_tokensOutput limit; lowered automatically when your credit can't cover it.
max_completion_tokensOutput limit (preferred over max_tokens when both are set); lowered automatically when your credit can't cover it.
temperaturetop_pnNumber of choices; each is reserved and billed.
seedpresence_penaltyfrequency_penaltyFunction tools. Provider-hosted tools (web search, code interpreter, MCP, …) are refused.
parallel_tool_calls{"type":"json_object"} or a json_schema.
reasoning_effortlow, medium or high on models with supports_reasoning.
logprobstop_logprobsPlayground chat completion › Responses
The completion (JSON) or an SSE stream.
idobjectcreatedmodelPlayground decision
Same as POST /v1/decisions, authenticated with the session and billed to the user.
Playground decision › Request Body
modelThe text to judge, or messages with input_text parts.
Questions of type predicate, choice (with choices) or score (with levels). Answers come back in the same order.
Playground decision › Responses
Answers.
model