Basic Memory
What's New

v0.21.0

Basic Memory Teams support — shared cloud workspaces that work consistently across MCP and CLI, plus faster search and sync and parsing footgun fixes in v0.21.

v0.21.0 brings Basic Memory Teams support: shared cloud workspaces now work consistently across every MCP tool and CLI command, so your AI assistant reaches the right projects whether you're working solo or with a team. Search and sync are also noticeably faster, and a handful of long-standing parsing footguns are gone.

Breaking change: relation parsing. Unquoted multi-word text before a wikilink is no longer treated as a custom relation type. If you rely on multi-word relation types, quote them before upgrading so they survive the next re-sync. See Upgrading below.

Basic Memory Teams Support

This is our first release to include support for Basic Memory Teams (subscription required). When you work with Teams, your projects live in a shared cloud workspace that everyone on the team can read and write. If you have a local workspace or personal Basic Memory subscription you can continue to use both at the same time.

What you'll notice:

  • Cross-workspace project discovery in MCP — your assistant's project list now spans every accessible cloud workspace, not just the current one.
  • bm project list spans all workspaces — no more missing projects depending on which workspace you were in.
  • Create projects by workspace slug from MCP — create_memory_project(workspace=...) targets a specific cloud workspace directly.
  • bm project add --visibility — set visibility for cloud projects at creation time.

Faster Search and Sync

  • Search hydration no longer issues an N+1 query — results come back faster, especially on large knowledge bases.
  • File indexing is batched during sync, and single-file sync is significantly quicker.
  • Vector sync does less redundant work, with tuned embedding defaults.

No configuration changes required — existing projects just get faster on upgrade.

Recent Activity, Fixed

recent_activity now orders and filters by updated_at, so notes you just edited bubble to the top instead of staying pinned to their creation time. Responses are also capped with an explicit truncation footer, so you can tell when there's more rather than silently losing rows.

Search Scope and Totals

  • Multi-project search is now opt-in. search_notes stays scoped to the resolved project by default; pass search_all_projects=True to search everything. This makes everyday search predictable.
  • Search responses include result totals, so paginated clients can stop guessing how many matches exist.

CLI Quality of Life

  • bm orphan surfaces entities whose underlying markdown files are gone, with a flag to clean them out.
  • bm db reset is now guarded — it refuses to run while a basic-memory mcp process is alive, so a reset can no longer corrupt an open session.
  • bm project info reports cloud index freshness; private projects show their display name instead of a raw UUID.

Training-Data-Friendly Tool Aliases

MCP tools now accept the parameter names models reach for naturally — q / search / text as aliases for query, and similar across other tools. Assistants spend fewer turns correcting argument names, so common operations just work on the first try.

Parsing Footgun Fixes

  • Obsidian callouts are skipped by the observation parser — > [!note] blocks no longer get misread as observations.
  • Long relation_type values are handled correctly (regression guarded).
  • The relation-parsing change below removes the biggest footgun of all: prose accidentally becoming relation types.

Upgrading

Relation parsing changed (#824). Unquoted multi-word text before a wikilink is no longer treated as a custom relation type:
  • Bare list wikilinks like - [[Target]] now index as links_to.
  • Prose list items like - some other thing [[Target]] now index as links_to.
  • Single-token relation types are unaffected: relates_to [[Target]] still works.
To preserve existing multi-word relation types across the next re-sync, quote them before upgrading: "relates to" [[Target]] or 'relates to' [[Target]].

The rest of the release is backward compatible — search, sync, and routing improvements apply automatically with no configuration changes.

Patch release: v0.21.1 is a CI-only release (Homebrew formula bump tooling) with no user-facing changes.

Documentation

For commit-level release history, see Changelog.

Getting Started

Configure where notes are saved and manage multiple projects.

Hermes Plugin

Give Hermes Agent persistent Basic Memory — search-before-answer recall, auto-capture, and 10 agent tools.