Server-Sent Events streaming for AtlasFlux responses - AtlasFlux API

Overview

AtlasFlux supports streaming for both /v1/responses and /v1/chat/completions. Enable streaming by setting "stream": true in your request body.

How streaming works

When streaming is enabled, the API returns a text/event-stream response with Server-Sent Events (SSE). Tokens are delivered incrementally as they are generated.

Chat Completions streaming

Uses OpenAI-compatible SSE format:

data: {"id":"chatcmpl_...","object":"chat.completion.chunk","choices":[{"delta":{"content":"Hello"}}]}
data: {"id":"chatcmpl_...","object":"chat.completion.chunk","choices":[{"delta":{"content":" world"}}]}
data: {"id":"chatcmpl_...","object":"chat.completion.chunk","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}
data: {"id":"chatcmpl_...","object":"chat.completion.chunk","choices":[],"usage":{"prompt_tokens":12,"completion_tokens":2,"reasoning_tokens":0,"cached_tokens":0,"total_tokens":14},"cost_myr":"0.000110"}
data: [DONE]

The terminal choice chunk always reports why generation ended through finish_reason. The following usage chunk reports settled cost as a six-decimal MYR string and separates visible completion tokens from reasoning tokens. If a chat stream fails after it has started, the stream emits a structured error before [DONE]:

data: {"type":"error","error":{"type":"provider_error","code":"provider_timeout","message":"AtlasFlux timed out while processing this request.","param":null,"request_id":"req_...","guidance":"Retry the request, reduce its size, or use streaming for longer responses.","action_url":null,"retryable":true}}
data: [DONE]

Responses streaming

Uses AtlasFlux-specific event types:

event: response.created
data: {"request_id":"arq_..."}

event: response.output_text.delta
data: {"delta":"The response text..."}

event: response.usage
data: {"usage":{"inputTokens":12,"outputTokens":48,"reasoningTokens":0,"cachedTokens":0,"costMicroMyr":1260},"request_id":"arq_..."}

event: response.completed
data: {"request_id":"arq_...","finish_reason":"stop"}

event: response.failed
data: {"error":{"type":"error","error":{"type":"server_error","code":"internal_error","message":"AtlasFlux could not complete the request due to an internal error.","param":null,"request_id":"req_...","guidance":"Retry with exponential backoff and contact AtlasFlux support with the request ID if it persists.","action_url":null,"retryable":true}},"request_id":"req_..."}

Implementation tips

Parse SSE events line by line
Handle event: and data: fields separately
Accumulate text deltas for the full response
Inspect finish_reason; length means the configured output-token limit was reached
Handle connection drops gracefully with retry logic
Handle response.failed and structured stream errors before closing the stream
The final event contains usage and cost information