Skip to content

v1.7.0

Latest
Latest

Choose a tag to compare

View all tags
@GabrielMartinMoran GabrielMartinMoran released this 27 Oct 01:18
v1.7.0
4e9a41e
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

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.

Assets 2
Loading