OpenAI SDK Compat

HealthyAudit Done
other
View Session

Health

Healthy

Benchmark

N/A

SDK Version

No SDK dependency

Audit Status

Audit Done

Started May 11, 2026

Approval

Not requested

Latest Audit Result

The OpenAI SDK compatibility guide (docs/reference/openai-sdk.mdx, live at docs.exa.ai/reference/openai-sdk) is accurate and current. It correctly documents three surfaces: (1) /chat/completions with model 'exa' routing to /answer, (2) /chat/completions with models 'exa-research'/'exa-research-pro' routing to /research, and (3) /responses with 'exa-research'/'exa-research-pro'. All endpoint URLs (https://api.exa.ai), auth patterns, model names, and extra_body usage match the backend code in Vulcan (chat-completion.ts, kronos.ts) and Kronos (research.controller.ts). The code examples are correct Python/JS/cURL with working streaming patterns. The /research/v1 endpoint was sunset May 1, 2026 per the changelog — the guide correctly uses /chat/completions and /responses instead, never referencing /research/v1 directly. No deprecated features are used in the guide. The 'exa-pro' model exists in the backend OpenAPI spec (exa-spec.yaml) but is not documented in the guide — this is intentional (undocumented/internal). The OpenAI Exa wrapper (exa.wrap) section and the Responses API tool-calling page are also consistent. Two minor updates and a few nice-to-have suggestions follow.

Health: healthy

Missing Features (1)

  • The create-a-task.mdx doc still references '/reference/chat-completions#research' as the link to the OpenAI compat interface, but that path 301-redirects to /reference/openai-sdk — the link works but the anchor #research is lost, so users land at the top of the page instead of the Research section