Skip to main content
Brime
Changelog

What we shipped.

New features, improvements, fixes, and breaking changes. Subscribe via RSS or follow @brime on Twitter.

  1. Feature

    Inline images now return as base64 or signed URLs

    When `inline_images=1` is set, the engine returns embedded images alongside markdown — either inline as base64 or as signed URLs you can fetch within 24 hours.

    • Pass `options.inline_images="1"` in the parse-init body to enable.
    • Failed image extractions surface as warning markdown instead of swallowing — you'll see exactly which pages couldn't be processed.
    • Anchors are emitted at the image top edge and inserted by Y position, so figures land in reading order rather than at document tail.
  2. Improvement

    Layout extraction v2 — better table boundaries, stable row order

    Rewritten layout pass treats columns and tables as first-class citizens. Multi-column documents extract in correct reading order and table cells preserve row alignment.

  3. Feature

    Brime API public beta is live

    Document Intelligence API is now generally available in beta. Free tier (100 pages/month) and Pro tier ($49/month, 10K pages) are open for self-serve signup.

    • Three regions on day one: us-east, sjc, fra.
    • Streaming progress events via Server-Sent Events on the parse-status endpoint.
    • Webhooks for completion notifications, with HMAC signature verification.
  4. Improvement

    Concurrency limits raised on Pro tier

    Pro accounts can now run up to 20 concurrent jobs (up from 5). Backpressure is automatic — overflow jobs queue with sub-second pickup.

  5. Breaking

    Breaking: parse-init body now requires `options` envelope

    Per-job options must be sent as `body.options.<key>` to parse-init rather than as query string parameters or top-level body keys. Existing query-string callers are now ignored — clients should migrate.

    • Affected keys: `inline_images`, `extract_tables`, `pages`.
    • Migration: move all per-job toggles into a single `options` object on the parse-init request body.
    • parse-start body.options is now ignored; parse-init is the single source of truth for job configuration.
  6. Feature

    OCR pipeline supports additional image formats

    JPG, PNG, and WebP attachments are now first-class inputs alongside PDFs. Same endpoint, same response shape — just point at any image.