Skip to content

v0.4.0 — Core Engine Integration + Time Decay + OpenClaw Plugin

Choose a tag to compare

View all tags
@Neboy72 Neboy72 released this 18 Jun 23:09
· 23 commits to main since this release
v0.4.0
d4fd00f

Nexus Memory v0.4.0

The biggest update yet: the MCP server is now connected to the full core engine. Graph, Auto-Discovery, Lifecycle, and Events are no longer dormant code — they're live.

🆕 New Features (June 19)

MCP Server → Core Engine Integration

  • SkillGraph initialization in MemoryStore
  • Auto-Discovery after every remember() — automatically finds related facts and creates graph edges (zero LLM cost, pure heuristics)
  • Lifecycle filtering in recall() — deprecated and rolled-back facts are excluded from results
  • Supersession in update() — updating a fact deprecates the old version, new version becomes canonical
  • Events on all operations — remember, update, and forget now fire audit events

Time Decay in Retrieval

  • Gauss-shaped score decay (offset=30 days, scale=365 days)
  • Recent memories rank higher, old ones fade gracefully
  • Only applies when timestamps are present — backwards compatible

PROCEDURE Memory Category

  • New MemoryCategory.PROCEDURE for workflow/procedural memory with step ordering
  • 7 categories total: fact, belief, session, rule, preference, procedure, temp

Staging with Real Embeddings

  • Replaced placeholder [0.0] * 512 vectors with actual embedding provider calls
  • Auto-detects and respects user-selected provider (Voyage, OpenAI, Google, Jina, Ollama, sentence-transformers)
  • Staging dimension now matches main collection dimension

📦 Also included (June 18)

  • OpenClaw native plugin — Auto-Recall + Auto-Capture via Qdrant REST
  • install_openclaw_plugin.sh — one-command install
  • 3-way architecture — Hermes Plugin + OpenClaw Plugin + MCP Server, same Qdrant collection

🔧 Changed

  • All version numbers synchronized (pyproject.toml, nexus.version, plugin.yaml, mcp_server.py)
  • ensure_collections() auto-detects vector dimension from embedding provider (was hardcoded 512d)
  • AGENTS.md categories updated to include procedure

🐛 Fixed

  • Staging placeholder vectors replaced with real embeddings
  • Personal data removed from public files
  • Test suite updated for new category count and dimension detection

📦 Install / Upgrade

# New install
git clone https://github.com/Neboy72/nexus-memory.git ~/nexus-memory
cd ~/nexus-memory && pip install -e .

# Upgrade existing
cd ~/nexus-memory && git pull && pip install -e .

No breaking changes — same Qdrant collection, same API, same tools. Lifecycle filtering is backwards compatible (entries without lifecycle_status pass through). Time decay only applies when timestamps are present.

🧪 Tests

389 tests passing. 3 pre-existing environment test failures (embedding provider detection edge cases when API keys are set in the test environment).

Assets 2
Loading