Skip to content

Release v3.9.0 - Global Memory#66

Merged
cdnsteve merged 9 commits intomainfrom
develop
Mar 17, 2026
Merged

Release v3.9.0 - Global Memory#66
cdnsteve merged 9 commits intomainfrom
develop

Conversation

@cdnsteve
Copy link
Copy Markdown
Member

Summary

  • Global memory layer for cross-project knowledge (~/.sugar/memory.db)
  • Project-first search strategy with reserved guideline slots
  • New guideline memory type for cross-project standards
  • --global flag on all CLI memory commands
  • scope parameter on MCP store_learning tool
  • MCP server works outside Sugar projects (global memory always available)
  • New sugar://global/guidelines MCP resource
  • Full docs and branding rewrite: "Persistent memory for AI coding agents"

Test plan

  • 65 new tests (981 total passing)
  • CI green across all platforms (Ubuntu, macOS, Windows) and Python 3.11-3.13
  • Manual CLI verification: remember --global, recall, memory-stats, forget
  • Cross-project recall verified (global results from outside Sugar projects)
  • Site deployed to sugar.roboticforce.io with updated branding

Add ~/.sugar/memory.db as a global memory store that works alongside
project-local memory. All search/recall operations now query both
stores and merge results by relevance.

- GlobalMemoryManager wraps project + global MemoryStore instances
- MCP server works outside Sugar projects (global-only mode)
- New --global CLI flag on remember/recall/memories/forget/memory-stats
- New "guideline" memory type for cross-project standards
- New sugar://global/guidelines MCP resource
- scope parameter on store_learning MCP tool
- Search results labeled with scope (project/global)
- 62 new tests, 978 total passing
… slots

Search now uses a tiered approach:
1. Project store searched first (local context always wins)
2. Global guidelines get reserved slots (always surface)
3. Remaining slots filled with other global results
4. Dedup across all results

This ensures project decisions override global ones, new projects
get global knowledge automatically, and cross-project standards
like SEO rules always appear regardless of local result count.
Update all documentation to lead with "Persistent memory for AI
coding agents" instead of task-queue-first messaging.

- README rewritten: memory capabilities lead, task queue secondary
- pyproject.toml description updated
- CLI help text updated
- Plugin README repositioned
- Quick start shows memory workflow before task management
- Memory docs: global memory section, guideline type, --global flag
- CLI reference: --global flag, scope labels, guideline type
- Installation guide: memory MCP as primary integration
- Docs index: memory system elevated to position 3
feat: Add global memory layer for cross-project knowledge
@cdnsteve cdnsteve merged commit b6e9f3c into main Mar 17, 2026
36 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant