Answer Endpoint Reference
Grounded answer generation surface via POST /answer.
Canonical Docs Links
- Base docs URL:
https://exa.ai/docs - Answer reference:
/reference/answer - OpenAI-compatible guide:
/integrations/openai-sdk
Contents
- Overview
- Request shape
- When to prefer
/answer - Structured output
- Streaming
- Response shape
- Critical pitfalls
Overview
Use the answer endpoint when you want Exa to:
- run the retrieval step
- generate a direct answer or concise summary
- attach citations in one API response
This is the best fit when the application wants a grounded answer, not just ranked results.
Request Shape
The recommended request is the query alone:
POST https://api.exa.ai/answer
{
"query": "What is the latest valuation of SpaceX?"
}Core Parameters
Everything beyond query requires an explicit task justification.
| Parameter | Type | Notes |
|---|---|---|
query |
string | Required |
stream |
boolean | Enables SSE token streaming |
text |
boolean | Include full text in cited source objects; adds payload size, so only when the task needs source text |
outputSchema |
object | JSON Schema for structured answers |
systemPrompt |
string | Guide answer behavior |
userLocation |
string | Two-letter ISO country code |
When To Prefer /answer
If the product already has a chat LLM, do not use /answer. Give that LLM /search as a tool instead (contents: { highlights: true }) so it owns generation and conversation context.
Use the answer endpoint when:
- you need a grounded answer with citations and Exa should do the generation
- you do not need to manually inspect or rank raw results first
- the app interface is question-driven rather than retrieval-driven
Prefer the search endpoint when:
- the product already has a chat LLM (expose
/searchas a tool) - your application needs the result list itself
- you want explicit control over
contentsmodes on each result - you want deeper synthesized search flows via
deep-liteordeep
Structured Output
from exa_py import Exa
exa = Exa(api_key="YOUR_EXA_API_KEY")
response = exa.answer(
"What is the latest valuation of SpaceX?",
output_schema={
"type": "object",
"properties": {
"valuation": {"type": "string"},
"date": {"type": "string"}
},
"required": ["valuation", "date"]
}
)
print(response.answer)Use small, explicit schemas. The answer endpoint is good for question-shaped structured output, while the search endpoint is often the better fit for broader synthesized retrieval.
Streaming
When stream: true, the answer endpoint returns server-sent events. In SDKs, use the dedicated streaming helpers instead of passing stream=True into ordinary non-streaming helper calls.
TypeScript:
for await (const chunk of exa.streamAnswer("Explain quantum computing")) {
if (chunk.content) {
process.stdout.write(chunk.content);
}
}Response Shape
Typical non-streaming response fields:
answercitationsrequestIdcostDollars
Each citation object may include URL, title, author, published date, and optionally text if requested.
Critical Pitfalls
- Use the answer endpoint for answer-shaped output, not as a substitute for full search result inspection.
- If you need rich per-result extraction controls, prefer the search endpoint.
- For SDK streaming, prefer
stream_answerorstreamAnswerhelpers over ad hoc raw flags.