Does AskVault's streaming query endpoint cost more than the standard one?

No. Both /v1/query and /v1/query/stream consume exactly 1 message per successful query regardless of response length.

Can I cancel an AskVault stream mid-flight?

Yes. Abort the fetch ReadableStream (or close the connection your client used). AskVault stops generating within 200 ms. Tokens generated up to that point are billed as 1 message.

How do AskVault skill-triggered responses behave in a stream?

Skill outputs are typically short and arrive as a small number of token chunks. Streaming still works but you may see just a few events rather than the steady token flow of a long answer.

POST /v1/query/stream endpoint

Written by Joseph Russo, Principal Architect, Networks and Tech, AskVault · Reviewed by Allison Chan

Last updated: June 18, 2026 · 3 min read

Endpoint

POST https://api.askvault.co/v1/query/stream

Same request body as /v1/query. Response uses Server-Sent Events (SSE).

Event format

Every line is a plain data: <json> frame — there are no named SSE event: types. Read the done boolean in each payload to tell a token frame from the final frame.

Example

curl -N -X POST https://api.askvault.co/v1/query/stream \
  -H "Authorization: Bearer ak_live_xxx" \
  -H "Content-Type: application/json" \
  -d '{"query":"What is your refund policy?"}'

Returns:

data: {"token":"Our ","done":false}

data: {"token":"refund ","done":false}

data: {"token":"policy ","done":false}

data: {"token":"","done":true,"sources":[{"document_name":"Refund Policy","url":"..."}],"tokens_used":450,"conversation_id":"conv_xxx"}

When to use streaming

Three cases:

Chat UIs. Show the bot's reply as it's typing, like ChatGPT.
Long responses. Reduce perceived latency.
Voice TTS streaming. Start speaking before the full answer is ready.

Latency

First token: about 500 ms (p50).
Full response: about 1.5 to 3 seconds (p50 to p95).

Streaming reduces perceived latency about 60 to 80%.

Client implementations

Browser: EventSource can't set an Authorization header, so it doesn't work directly against this endpoint — use fetch with a ReadableStream instead (see the full streaming reference for a worked example), or proxy through your backend.

const res = await fetch('/v1/query/stream', { method: 'POST', headers: {/* Authorization, Content-Type */}, body: /* ... */ });
const reader = res.body.getReader();
const decoder = new TextDecoder();
// read chunks, split on "\n\n", parse each "data: {...}" line, check .done

Node.js: Native fetch with a ReadableStream, or a streaming-aware HTTP client.

Python: requests with stream=True and iter_lines(), or httpx streaming.

Error handling

Mid-stream errors arrive as a normal data: frame with done: true:

data: {"error":"Rate limit exceeded","done":true}

Client should close the connection and retry with backoff.

Limits

Same as /v1/query. Auth, rate limits, query length.
Stream duration cap. 30 seconds before forced close.
Reconnect. No native reconnect; client retries with new request.

Common pitfalls

Browser blocks the connection. CORS plus credentials issue. Confirm Authorization header allowed.

Stream stalls. Network buffering. Use flush() on your client framework.

Lost tokens. Network drop mid-stream. Implement client-side retry from last-known position.

FAQ

Does streaming cost more than standard?

No. Same per-query billing.

Can I cancel a stream mid-flight?

Yes, by closing the connection (e.g. aborting the fetch ReadableStream). Server stops generating; tokens billed up to that point.

Does streaming work for skill-triggered responses?

Skill output is short and arrives in batches. Streaming still works but may show as fewer chunks.

Was this page helpful?