Changelog — bRRAIn Docs

Brain releases of bRRAIn, newest first, taken from the published release notes.

Changelog

This page lists brain releases (the brrain service an organization's vault runs on), newest first. Each entry is taken from the release notes published with that deploy. Console, Nexus, extension and website releases have their own notes and are not repeated here.

To see what a host is running, request /version. To see what is available, run brrain upgrade -check. See Updates & releases.

1.2.82 — 2026-09-13

  • Vault files can be read as exact bytes (GET /api/vault/file?raw=1), with range requests and the same access checks as before. PDFs, images and Office files opened from the vault no longer come back corrupted or blank. HTML, SVG and JavaScript are served as plain text so a vault file never executes.
  • In-place upgrade fixes: the brain now restarts itself on the new release after an in-place upgrade, only signals the server for its own install root, and records the upgraded version so a hosted pod does not fall back to its original pin on the next container restart.
  • 1.2.81 was published and superseded by 1.2.82 the same day.

1.2.80 — 2026-09-13

  • Search finds PDF, DOCX, PPTX and XLSX content. Text extraction from Word and PowerPoint files was fixed, damaged archives are salvaged where possible, and partial text is kept from a damaged part.
  • At startup the brain extracts text for binary files that never had it, or had it from an older extractor. Disable with BRRAIN_EXTRACT_BACKFILL=off.

1.2.77 — 2026-09-05

  • First release carrying brrain upgrade: replace the whole release in place with SHA-256 verification, keep the previous release for -rollback, and leave extensions and the model store untouched.
  • The installer distinguishes a pinned version from the latest release: a hosted pod installs a version it was explicitly pinned to and does not silently follow latest.json.

1.2.74 — 2026-07-10

  • The vault's agents and advisors list is served from a cache refreshed in the background, so the console Agents page loads quickly on large vaults. TTL: BRRAIN_AGENTS_CACHE_TTL_SECONDS (default 120).

1.2.73 — 2026-07-10

  • Vault search is backed by an in-memory full-text index instead of scanning every file per query, with a fallback to the scan while the index builds. Binary documents are searchable through their extracted text and are reported under their own paths. Access rules apply to every hit as before. TTL: BRRAIN_SEARCH_INDEX_TTL_SECONDS (default 300).

1.2.70 — 2026-06-30

  • The ontology graph is served from a cache refreshed in the background and invalidated on every vault write, so the console graph renders on large vaults instead of timing out. Each visibility level gets its own cached view. TTL: BRRAIN_ONTOLOGY_CACHE_TTL_SECONDS (default 120).

1.2.69 — 2026-06-29

  • Backing out a vault import runs in the background, so a long back-out is not cut off by a dropped connection. Any past import can be backed out from the console's Recent imports list.

1.2.68 — 2026-06-29

  • Vault import can store binary files verbatim, place files at the vault root, and map each source folder to a chosen vault destination. Path safety checks and write permissions still apply.

1.2.67 — 2026-06-26

  • Vault import from a GitHub repository, run from the console: preview, import with skip-if-exists or merge, an import manifest, integrity checks, and a full back-out.

1.2.66 — 2026-06-16

  • MCP tool fabric for Exchange MCP servers, so MCP servers installed from the Exchange are usable through every connected AI client.

1.2.64 — 2026-06-14

  • MCP clients discover the brain's tools dynamically, so new tools reach Claude Desktop, VS Code and Claude Code without reinstalling the client. Data Pipe retrieval is available as an MCP tool.

1.2.62 — 2026-06-05

  • Sensitive vault routes require a verified identity token from the console and reject requests that only assert a role in a header.