v1.7.0

mind v1.7.0

Mind reaches top-three territory by raw score in an indicative comparison
with published OmniMemEval LongMemEval-S results. This substantial update pairs
that result with more capable search, a renewed web app, and stronger
cross-session memory continuity.

Highlights

• More capable search across every surface — Mind combines full-text,
  phrase, substring, typo-tolerant, short-term, and optional semantic retrieval
  into one deterministic ranking. Accent-insensitive matching and improved
  recall for natural-language questions help surface useful partial matches.
  CLI, MCP, HTTP API, and web search share the same ranking and confidence
  signals.

• A renewed web experience — The redesigned app brings search and shared
  filters into each space, with confidence and semantic controls, multi-tier
  filtering, clearer logs, and URL-restored views. The accessible Neural Map
  makes it easier to explore how memories connect.

• More reliable memory continuity — Checkpoint completion now creates the
  session summary, preserves directional links, and removes the checkpoint
  atomically. Date filters behave consistently, and memory links retain their
  labels and direction.

• Simpler agent setup — mind setup configures supported integrations and
  installs the shared mind-management skill; use mind setup <agent> to
  configure one agent. Pi Agent is now supported.

• OmniMemEval result — Mind's best completed lexical-only run scored
  86.67% answer accuracy. The comparison is indicative, not an official
  ranking: Mind used a stratified subset and GPT-6 Luna (low) for both answering
  and judging, while published results use 500 questions and different
  evaluation settings. See the benchmark methodology and results.

Search compatibility

• Only uppercase AND, OR, and NOT are boolean operators.
• A supplied search with no informative terms now returns a validation error.
  Omitting the search field retains its listing behavior.
• MCP memory_query search responses are compact by default. Set
  USE_DEBUGGING_EXPANDED_TOOLS_RESPONSE=true in the Mind MCP process to
  include detailed score and method diagnostics.

Upgrade notes

mind update

mind update refreshes detected agent integrations automatically. Restart
your agent host after updating so it reloads Mind's MCP process and refreshed
integration files. Use mind update --no-refresh-integrations to defer that
refresh.

v1.6.1

mind v1.6.1

Patch release focused on the OpenCode automation plugin: full OpenCode V2
support while keeping V1 compatibility, correct project-space resolution,
quieter and bounded plugin state, and accurate session identities on both
runtimes.

Highlights

• OpenCode V2 support — the generated ~/.config/opencode/plugins/mind-automation.js
  now default-exports a V2 plugin definition (id + setup) with native
  context/compaction session hooks and ctx.event.subscribe, while keeping
  the V1 server() hooks and the named export. V2 no longer rejects the plugin
  with Plugin must export a default definition…, and existing V1 installs keep
  working (verified on V1 1.18.28/1.18.33 and V2 2.0.19).
• One project space per canonical project — continuity artifacts are stored
  under projects/<canonical-project> even when events come from a nested or
  different directory.
• Quieter, bounded plugin state — state is written only after
  state-changing events, and the checkpoint/summary/handled maps are capped to
  the newest 400 entries.
• Accurate session identities — the plugin resolves the real session id on
  V2 stream events (data.sessionID) and V1 bus events
  (properties.sessionID / properties.info.id) instead of the event id, so
  dedupe keys and session summary names are correct on both.

Fixed

• OpenCode automation plugin now loads on OpenCode V2: default export with
  id + setup, native V2 session hooks, and ctx.event.subscribe, keeping
  V1 server() hooks for compatibility.
• V2 automation resolves one project space per canonical project path
  (location.project.canonical takes precedence over the event directory).
• V2 automation writes plugin state only on state-changing events; state maps
  are capped to the newest 400 entries; empty or non-string session id fields
  no longer mask later fallbacks.
• V1 automation now uses the session id from event properties
  (properties.sessionID, properties.session.id, properties.info.id)
  instead of the bus event id, so checkpoint dedupe keys and session summary
  names use the real session.

Upgrade notes

mind update

mind update refreshes detected agent integrations automatically, including
regenerating ~/.config/opencode/plugins/mind-automation.js. No manual
mind setup refresh is required. Use --no-refresh-integrations if you need
to defer the refresh.

Restart your agentic host after updating — fully quit and relaunch
OpenCode (and any other agent host) so the refreshed plugin file and the
updated mind mcp stdio process are re-read from disk. Running hosts keep
the previous plugin and protocol cached until restarted.

v1.6.0

mind v1.6.0

Minor release focused on sharper agent memory discipline and a couple of
reliability fixes. Agents now get clearer rules for what to persist as durable
memory, a complete living reference taxonomy, and MCP search is no longer
vulnerable to a fallback bug.

Highlights

• Sharper durable-memory guidance — agents now follow a four-point
  pre-creation checklist (future utility, novelty, evidence, stability) before
  writing a durable memory, reducing noise from session-summary-style content.
• Living reference taxonomy formalized — 7 reference types
  (ref:project-map, ref:architecture, ref:domain, ref:workflow,
  ref:style, ref:key-decisions, ref:known-pitfalls) with explicit
  creation/update triggers, content format, and 30-day maintenance reviews.
• MCP memory_query no longer crashes on the LIKE fallback when FTS5
  returns no results.
• OpenCode session-end automation is ~95% quieter — the automation plugin
  now only persists a session summary once per session instead of on every idle.

Changed

• Memory protocol guidance refined: durable memories now require passing a
  four-point checklist (future utility, novelty, evidence, stability) before
  creation.
• Living reference taxonomy formalized with 7 reference types, creation
  triggers, update triggers, content format, and 30-day anti-staleness reviews.

Fixed

• MCP memory_query no longer fails with no such column: s.name when FTS5
  returns no results and the LIKE fallback is used.
• OpenCode automation plugin reduced session summary noise by ~95% by
  switching from session.idle (20-50 fires/session) to session.deleted
  (1 fire/session).

Upgrade notes

mind update

mind update automatically runs the integration refresh for detected agent
integrations (OpenCode, Claude Code, Codex, Cursor, etc.). No manual
mind setup refresh is required. After the update, agents will pick up the
new memory protocol guidance automatically through mind mcp on the next
system_instructions call. Use --no-refresh-integrations on mind update
if you need to defer the integration refresh.

Restart your agentic system after installing — fully quit and relaunch
the editor/agent host (OpenCode, Claude Code, Codex, Cursor, etc.) so the
updated mind mcp stdio process and refreshed integration configs are
re-read from disk. In-session MCP servers keep the previous binary and
protocol files cached, so a hot reload won't pick up the new guidance
until the host is restarted.

v1.5.1

mind v1.5.1

Patch release with a fix for the Neural Map graph view in the web UI.

Highlights

• Neural Map rendering fixed — graph nodes now position correctly around
  their tier rings instead of clustering at the upper-left corner of the canvas.

Fixed

• Nodes on the Neural Map collapsed to the upper-left origin when a space
  used UUID memory identifiers (default since v1.5.0 schema migration).
  The layout engine assumed numeric IDs and produced NaN coordinates;
  it now uses deterministic string-aware sorting and hashing so nodes
  render at the correct tier-ring positions regardless of ID format.

v1.5.0

mind v1.5.0

This release makes mind easier to keep in sync, safer to update, and more
reliable for agent workflows.

Highlights

• Added experimental autosync commands to mirror spaces into versioned .mind/
  files for easier review, sharing, and recovery.
• Added safer database migrations with verified backups, automatic restore on
  migration failure, and backup retention.
• Added mind setup refresh so existing agent integrations can stay aligned
  after updates without recreating unrelated configs.
• Moved session summaries into projects/<repo> as same-space session-*
  memories and added a migration command for older summaries.
• Added UUID primary IDs for spaces and memories while preserving public
  name-based CLI and MCP contracts.
• Added safer setup config writes for JSON, JSONC, and TOML agent configuration
  files.
• Fixed the public installer launcher path so new installs run the current
  src/mind.ts entry point.
• Improved mind update so it validates migrations before refreshing
  integrations.

Added

• Added autosync commands that mirror spaces into versioned .mind/ files.
• Added mind setup refresh for refreshing detected mind-managed integrations.
• Added mind migrate sessions <project-space> [--dry-run] for migrating older
  sessions/<repo> summaries into the project space.
• Added maintainer release automation support for curated GitHub release notes
  with --notes-file <path> and package-lock version alignment.
• Added OpenCode opencode.jsonc support, including comment-preserving edits
  for existing JSONC configuration.

Changed

• Changed database migrations to create verified backups, restore automatically
  on failure, and keep the latest automatic backups.
• Changed mind update to validate migrations before refreshing integrations.
• Changed session summaries to live in projects/<repo> instead of
  sessions/<repo>.
• Changed internal IDs for spaces and memories from integer autoincrement IDs to
  UUID text primary keys, with memories.fts_id retained as the SQLite FTS5
  surrogate.

Fixed

• Fixed a critical data-loss risk in mind setup and mind setup refresh where
  malformed JSON or JSONC config files could be overwritten with empty defaults.
• Fixed Codex TOML setup handling so malformed TOML aborts safely and existing
  unrelated tables and comments are preserved.
• Fixed the public installer launcher so new installs run the current entry
  point.
• Fixed protocol wording and update flows that could confuse setup, migration,
  or session recovery workflows.

Upgrade notes

Run:

mind update
mind setup refresh

If you have older session summaries under sessions/<repo>, run:

mind migrate sessions projects/<repo> --dry-run
mind migrate sessions projects/<repo>

Review the dry run before applying the migration.

v1.4.0

What's Changed

• Develop by @GabrielMartinMoran in #3

Full Changelog: v1.3.0...v1.4.0

v1.3.0

What's Changed

• Feat/enhancements by @GabrielMartinMoran in #1

New Contributors

• @GabrielMartinMoran made their first contribution in #1

Full Changelog: v1.2.1...v1.3.0

v1.2.1

Full Changelog: v1.2.0...v1.2.1

v1.2.0

Full Changelog: v1.1.1...v1.2.0

v1.1.1

Full Changelog: v1.1.0...v1.1.1