Also in this release
MemHub + APIFeatures12
- Brains get a generated overview, refreshed nightly and debounced against content changes.
- A per-brain activity log, exposed at
GET /v2/cb-activity and documented in OpenAPI. - Skills can carry binary brand assets — logo, font, template — inside the bundle.
- REST CRUD for team skills at
/v1/team/skills. - Collection CRUD and the developer surface are ported into MemHub at
/v1/developer. - MCP server —
search_brains routes across brains over manifest embeddings. - MCP server —
search_all_brains searches items across every brain you can reach. - MCP server —
get_brain_overview reads a brain's generated overview. - MCP server —
delete_skill, plus a refresh mode on add_skill_to_brain. - MCP server — edit access for agent brains: share upsert and listing.
- MCP server — Linear, Apollo.io, Calendly, and Ahrefs join the connectable providers.
- APINative
GET /mem-manager/usage/summary, and self-serve management for the API subscription.
Improvements4
- OCR'd documents are chunked one chunk per page, so citations point at the right page.
- Agent brains show who shared them and when.
- Personal orgs are retired; MemHub is team-only.
- APIBreaking: the directive scope field was renamed
scope_repo → scope, then wired through as namespace on the API surface.
Fixes4
- Chat — files produced inside a turn resolve when saving; saving works in brain-less chats; NUL bytes no longer break persisted tool payloads.
- Notion index — in-flight builds resume after a worker dies and cancel cleanly on shutdown, instead of running concurrently.
- Brain digest — object-shaped decision items render as prose, not raw JSON.
- APIEmbeddings — truncation is bounded on UTF-8 bytes rather than characters.