Migrating from Exa

Brime exposes Exa's wire format under /exa/v1/*. Three endpoints supported (/search, /answer, /contents) plus streaming for search and answer.

One-line migration

from exa_py import Exa

client = Exa(
    api_key="sk-brime-...",
    base_url="https://api.brime.dev/exa",
)

result = client.search("Tesla Cybertruck recall 2025")

import Exa from "exa-js";

const exa = new Exa("sk-brime-...", { baseUrl: "https://api.brime.dev/exa" });
const result = await exa.search("Tesla Cybertruck recall 2025");

What's compatible

Exa endpoint	Brime path	Status
`/search`	`/exa/v1/search`	✅ Full (synthesised when `type=neural/deep*`)
`/answer`	`/exa/v1/answer`	✅ Full
`/contents`	`/exa/v1/contents`	✅ Full
`/findSimilar`	—	❌ Vector retrieval not yet (engine work)
`/research/v0/tasks`	—	❌ Not yet

stream: true is supported on /search and /answer — frames come as OpenAI chat.completion.chunk envelopes.

Field mapping

Exa field	Brime behaviour
`type: neural \| deep \| deep-lite \| deep-reasoning \| instant`	All map to Brime ranking (vector retrieval is on the roadmap)
`category` (people / company / paper / news / personal / financial)	Silently ignored
`userLocation`	Silently ignored (no per-country routing)
`additionalQueries`	Silently ignored
`outputSchema`	Silently ignored (no structured output yet)
`useAutoprompt`	Silently ignored (no query expansion)
`contents.highlights`	Always `[]`
`contents.summary`	Always `""`
`contents.subpages`	Always `[]`
`contents.livecrawl + livecrawlTimeout`	Silently ignored
`costDollars`	Synthesised from Brime credits (rough estimate)

Hit a parity gap that's not listed? Open an issue at github.com/orgs/brime-dev/repositories.

Authentication

Both Authorization: Bearer <key> and Exa's x-api-key: <key> header work. Use whichever your client sets by default.

Streaming

Exa chat.completion.chunk envelope is preserved:

data: {"id":"...","choices":[{"delta":{"content":"The ECB held rates at 3.25% in October 2025..."}}]}
data: {"id":"...","citations":[{"url":"https://en.wikipedia.org/wiki/Generative_artificial_intelligence"}]}
data: [DONE]

The streaming layer is synthesised — Brime runs the underlying call synchronously, then chunks the response into SSE frames. Real LLM token-level streaming is on the roadmap.

Pricing

Brime credits. See Pricing.

When to use the Exa adapter vs the native `brime` SDK

Use the Exa adapter to keep an existing Exa codebase running with minimum churn.
Use the native brime SDK when starting fresh — wider feature surface (deep research, SSE replay, idempotency).