Skip to content

Release Notes

Jump to bottom
Bellok edited this page Apr 3, 2026 · 7 revisions

Release Notes

Back to Home

Release history for the Neurodivergent Memory MCP Server. All 0.x releases are research previews.

Unreleased — current v0.3.0 development snapshot

Release-readiness snapshot as of 2026-04-03: the development branch has the planned v0.3.0 features merged, but they are not yet published as a tagged npm or container release.

Delivered on the active branch

  • distill_memory for structured emotional-to-logical translation
  • Behavioral loop guardrails on top of telemetry, including no-net-new-info warnings and distillation suggestions
  • agent_id attribution on writes plus per-agent breakdowns in memory_stats
  • Goal-aware retrieval via optional context, recency_weight, and intensity filtering
  • LUCA-addressed custom districts via register_district
  • Import and storage diagnostics improvements including storage_diagnostics, file-based import, dry_run, dedupe policies, and snapshot migration controls
  • Epistemic-status consistency improvements: practical_execution task memories now default to draft when omitted, and related_to plus list_memories now accept epistemic_statuses filters alongside search_memories
  • Packetized synthesis prompt support and tool mirrors for prompt-derived context

Publication status

  • Latest published npm package: 0.2.0
  • Latest published Docker tags: 0.2.0
  • Current v0.3.0 work should be treated as branch state until an explicit release is cut

v0.2.0 — 2026-04-01

Minor release — Trust & Telemetry: WAL persistence hardening, concurrency safety, structured MCP error contract, loop telemetry, performance benchmarks, and project_id scoped retrieval.

Added

  • WAL (write-ahead log) persistence with crash recovery, replay on startup, and configurable eviction
  • Async write mutex and write queue with backpressure to prevent data corruption under concurrent tool calls
  • Structured MCP error contract: stable error codes (NM_E001–NM_E0xx), human messages, and recovery hints on every failure path
  • Observe-only loop telemetry: repetition counters, similarity scoring across recent writes, and ping-pong detection between districts
  • Performance benchmark suite: store throughput and BM25 search latency at 1k / 5k / 10k memories
  • project_id field on store_memory and scoped retrieval via retrieve_memory and search_memories
  • RC release channel: development branch published as a pre-release channel for early testing
  • Memory-driven issue execution prompt for agent instructions

Fixed

  • Docker persistence path resolution (volumes now correctly mount to the configured storage path)
  • Deterministic project_id in live smoke tests; immutable RC tag handling in release workflow
  • Resolved open Copilot review threads from PR #29 (WAL hardening follow-ups)

Performance Baseline (Published)

Scale store p95 search p95
1k memories <10 ms <5 ms
5k memories <25 ms <15 ms
10k memories <60 ms <30 ms

Known Limitations

  • Telemetry is observe-only in the published release; behavioral guardrails now exist on the active v0.3.0 development branch
  • Custom districts are not in the published release; LUCA-addressed custom districts now exist on the active v0.3.0 development branch
  • agent_id attribution is not in the published release; it now exists on the active v0.3.0 development branch

v0.1.8 — 2026-03-28

Patch release: Forces Docker Hub README refresh on each release and upgrades Node to 24.

Added

  • Added peter-evans/dockerhub-description@v4 step in the release workflow so Docker Hub description is always updated from README.md on every release
  • Added workflow_dispatch to allow manual README refresh runs when needed

Changed

  • Upgraded workflow Node runtime from 20 to 24
  • Upgraded Docker build/runtime base images from node:20-alpine to node:24-alpine

v0.1.7 — 2026-03-28

Patch release: Fixes README GIF and link rendering on Docker Hub and GHCR.

Fixed

  • Replaced relative GIF paths with absolute raw.githubusercontent.com URLs so Docker Hub and GHCR can render the preview image and link correctly

v0.1.6 — 2026-03-28

Patch release: Disables Docker latest tag generation to comply with immutable tag settings.

Fixed

  • Release workflow now sets Docker metadata flavor.latest=false to avoid pushing immutable latest tag
  • Prevents Docker Hub publish failures caused by immutable floating tags

v0.1.5 — 2026-03-28

Patch release: Fixes Docker Hub immutable tag conflict and publishes to the official MCP Registry.

Added

  • Added mcpName field to package.json required for MCP Registry namespace verification (io.github.jmeyer1980/neurodivergent-memory)
  • Added server.json for publishing to the official MCP Registry at registry.modelcontextprotocol.io

v0.1.4 — 2026-03-28

Patch release: Release pipeline hardening and rerun safety.

Fixed

  • Release workflow now syncs release branch with main on version tag pushes
  • Release workflow now skips npm publish when the version already exists on npm (rerun-safe)
  • Continued hardening for GitHub Actions JavaScript runtime migration to Node 24

v0.1.3 — 2026-03-28

First fully successful multi-channel release in this project lifecycle.

What Changed

Improves package visibility in GitHub by publishing container images to GitHub Container Registry (GHCR) in addition to Docker Hub.

Fixed

  • Release workflow now publishes container images to ghcr.io/jmeyer1980/neurodivergent-memory
  • Added packages: write workflow permissions required for GHCR publishing
  • GitHub Packages listing now available through GHCR package pages linked to this repository

v0.1.2 — 2026-03-28

Patch release: Release pipeline reliability and publish resiliency.

Fixed

  • Release workflow now syncs release with main on version tag pushes
  • Release workflow now skips npm publish when the version already exists on npm (rerun-safe)
  • Continued hardening for GitHub Actions JavaScript runtime migration to Node 24

v0.1.1 — 2026-03-28

Initial research preview release.

Added

  • Core MCP Server — knowledge graph storage and retrieval via Model Context Protocol stdio transport
  • 5-District Architecture — semantic organization across logical_analysis, emotional_processing, practical_execution, vigilant_monitoring, and creative_synthesis domains
  • BM25 Full-Text Search — relevance-ranked memory retrieval with configurable scoring and filtering
  • Graph Connections — bidirectional edges between memory nodes for associative traversal
  • Canonical Tag Schema — enforced tagging (topic, scope, kind, layer) for structured knowledge organization
  • Persistence — JSON-based durable storage at .neurodivergent-memory/memories.json
  • Tools: store_memory, retrieve_memory, search_memories, delete_memory, connect_memories, traverse_from, related_to, list_memories, memory_stats, import_memories
  • Docker Support — multi-platform container images (amd64, arm64) with provenance and SBOM attestation
  • Security — npm provenance publishing, container attestation, checksums on release artifacts

Fixed

  • Test script hang prevention — added adaptive timeout and completion detection to prevent indefinite stdin blocking

Known Limitations

  • Single-file persistence — no journal/WAL pattern; risk of data loss on unclean shutdown (resolved in v0.2.0)
  • No concurrency control — unsafe for simultaneous writes from multiple agents (resolved in v0.2.0)
  • Static districts — cannot adapt semantic organization per use case
  • Basic query language — BM25 ranking ignores agent goals or context
  • No observability — no logging, metrics, or debug endpoints (partially resolved in v0.2.0)
  • No resource controls — unbounded memory growth potential

See also: Architecture · Getting-Started · Roadmap

Home

Setup

Reference

Reports

v0.3.0 · Research Preview

npm · Docker Hub · GHCR

Clone this wiki locally