Skip to content

Releases: runtimenoteslabs/memory-layer

v2.1.1 - Bug Fixes

24 Jan 07:08
@AnandDN AnandDN
v2.1.1
e3b90b5

Choose a tag to compare

View all tags
v2.1.1 - Bug Fixes Latest
Latest

Bug Fixes

  • Fixed Tasks API routes - /tasks/stats and /tasks/context now
    work correctly (route ordering issue)
  • Fixed plugin config - Correct version number and repository URL
    in .claude-plugin/plugin.json
  • Fixed Web UI - Header content now centered to match main content
    area

Cleanup

  • Simplified .mcp.json - MCP server advertises tools dynamically
  • Removed docs/MIGRATION_GUIDE.md - No public v1 exists to migrate
    from

Installation

pip install
git+https://github.com/runtimenoteslabs/memory-layer.git@v2.1.1
Assets 2
Loading

v2.1.0 - Claude Code Tasks Integration & Web UI

24 Jan 06:53
@AnandDN AnandDN
v2.1.0
7764868

Choose a tag to compare

View all tags
v2.1.0 - Claude Code Tasks Integration & Web UI

Memory Layer v2.1.0

Persistent memory for AI coding agents with outcome-based learning.

What's New

Claude Code Tasks Integration

  • Read tasks from ~/.claude/todos/ (native Claude Code task system)
  • Unified Tasks API combining Beads + Claude Code sources
  • Automatic outcome recording when tasks complete
  • CLI: mem tasks, mem tasks-sync, mem tasks-context, mem tasks-stats
  • REST API: /tasks endpoints with source filtering
  • MCP: tasks_list, tasks_sync, tasks_context, tasks_stats tools

Web UI at localhost:8080

  • Dashboard with color-coded category bars (all 10 categories)
  • Memories list with filtering (category, project, search)
  • Search with semantic and keyword modes
  • Tasks tab with unified Beads + Claude Code view
  • Memory detail modal with outcome recording
  • Light/dark theme toggle
  • Export to JSON

Hardening

  • Custom exception hierarchy with user-friendly messages
  • Graceful shutdown on single Ctrl+C
  • Configurable log levels (-v for INFO, -vv for DEBUG)
  • --full flag for complete memory content in list/search
  • mem check command for health verification

Highlights

  • 5-signal hybrid retrieval: Semantic (35%) + Outcome (25%) + Recency (15%) + Frequency (15%) + Confidence (10%)
  • Asymmetric outcome scoring: +0.2 for worked, -0.3 for failed (bad advice sinks fast)
  • 1,184 tests passing
  • Local-first: SQLite at ~/.memory-layer/memories.db

Installation

pip install git+https://github.com/runtimenoteslabs/memory-layer.git

Quick Start

# Add a memory
mem add "Always validate JWT server-side" -c convention

# Search
mem search "JWT"

# Start Web UI
mem serve --rest --port 8080

# Start MCP server (for Cursor, OpenCode)
mem serve --mcp

Documentation

- README.md - Overview and quick start
- USER_GUIDE.md - Detailed usage guide

What's Next

- v2.2: Reflection synthesis (periodic pattern analysis)
- v2.3: Team collaboration features
Loading