Skip to content

V0.5.0: feat: add Procedural Memory and Triggers pages with API integration

Choose a tag to compare

View all tags
@rexdivakar rexdivakar released this 11 Feb 20:15
· 22 commits to main since this release
V0.5.0
ca24195

User description

Summary

HippocampAI v0.5.0 — 6 major new intelligent memory features, comprehensive documentation overhaul, and full React frontend.

New Features (v0.5.0)

  • Real-Time Incremental Knowledge Graph — Auto-extraction of entities, facts, relationships on every remember() call using NetworkX in-memory graph with JSON persistence
  • Graph-Aware Retrieval — 3-way Reciprocal Rank Fusion (vector + BM25 + graph) with new GRAPH_HYBRID search mode
  • Memory Relevance Feedback Loop — User feedback (relevant/not_relevant/partially_relevant/outdated) with exponential decay scoring
  • Memory Triggers / Event-Driven Actions — Configurable webhooks, websocket, and log actions on memory lifecycle events (on_remember, on_recall, on_update, on_delete, on_conflict, on_expire)
  • Procedural Memory / Prompt Self-Optimization — Rule extraction from interactions, rule injection into prompts, effectiveness tracking via EMA
  • Embedding Model Migration — Safe background re-encoding via Celery with progress tracking

Also Included (v0.4.0 features)

  • Sleep Phase / Active Consolidation with DuckDB persistence
  • Multi-Agent Collaboration with shared memory spaces
  • React Frontend with analytics, graph view, health monitoring
  • Predictive Analytics, Auto-Healing Pipeline
  • Bi-Temporal Facts, Context Assembly, Custom Schemas, Benchmarks
  • Plugin System, Namespaces, Export/Import, Offline Mode, Tiered Storage
  • LangChain & LlamaIndex adapters
  • Audit Logging & Usage Tracking

Infrastructure

  • 242 files changed, +56,560 / -5,480 lines
  • 16 new config fields, 15 new REST API endpoints, 2 new Celery tasks
  • 6-weight score fusion: sim + rerank + recency + importance + graph + feedback
  • Package restructure: hippocampai.core (library) + hippocampai.platform (SaaS)

Documentation

  • Full rewrite of docs/CONFIGURATION.md
  • New: docs/COMPETITIVE_ADVANTAGES.md (vs mem0, Zep, Letta, Cognee, LangMem)
  • Updated: FEATURES.md, API_REFERENCE.md, CHANGELOG.md, CELERY_GUIDE.md, README.md
  • 6 new feature sections with architecture diagrams and code examples

Code Quality

  • pyright: 0 errors
  • ruff: all checks passed
  • Tests: 545 passing
  • No breaking changes — all existing APIs remain backward-compatible

Test plan

  • pyright src/ — 0 errors
  • ruff check src/ — clean
  • pytest — 545 tests passing
  • Docker compose up with all services
  • Verify new API endpoints via Swagger UI at /docs
  • Test knowledge graph extraction on remember() calls
  • Test 3-way RRF retrieval with ENABLE_GRAPH_RETRIEVAL=true
  • Verify feedback, triggers, procedural memory, and migration endpoints

PR Type

Enhancement, Bug fix, Tests, Documentation

Description

HippocampAI v0.5.0 Release — Comprehensive intelligent memory system with 6 major new features, full documentation overhaul, and production-ready implementation.

Core Features Added:

  • Real-Time Incremental Knowledge Graph — Auto-extraction of entities, facts, and relationships on every remember() call using NetworkX with JSON persistence

  • Graph-Aware Retrieval — 3-way Reciprocal Rank Fusion (vector + BM25 + graph) with new GRAPH_HYBRID search mode

  • Memory Relevance Feedback Loop — User feedback system (relevant/not_relevant/partially_relevant/outdated) with exponential decay scoring

  • Memory Triggers / Event-Driven Actions — Configurable webhooks, websocket, and log actions on memory lifecycle events (on_remember, on_recall, on_update, on_delete, on_conflict, on_expire)

  • Procedural Memory / Prompt Self-Optimization — Rule extraction from interactions, rule injection into prompts, effectiveness tracking via EMA

  • Embedding Model Migration — Safe background re-encoding via Celery with progress tracking

Implementation Highlights:

  • 6 new feature managers integrated into MemoryClient: graph_retriever, feedback_manager, trigger_manager, procedural, migration_manager

  • 15 new REST API endpoints across 4 new route modules (feedback, triggers, procedural, migration)

  • 2 new Celery background tasks for embedding migration and procedural rule consolidation

  • 16 new configuration fields for feature control and tuning

  • Extended score fusion to 6-weight model: similarity + rerank + recency + importance + graph + feedback

  • Knowledge graph persistence via JSON serialization with auto-save intervals

Bug Fixes & Robustness:

  • Fixed NetworkX graph degree calculations for compatibility

  • Fixed Qdrant API type compatibility (VectorParams, PointIdsList, Filter conditions)

  • Added null safety checks throughout (session_id, payload, task_id, trace.end_time)

  • Fixed memory attribute references (timestamp → created_at)

  • Improved benchmark runner error handling and retry logic

  • Fixed UUID conversion in middleware auth checks

API Refactoring:

  • Refactored user_id parameter passing: now passed per-call instead of in client constructor

  • Updated method signatures: track_session_message() uses text/user_id, removed deprecated batch methods

  • Fixed embedder parameter naming (model → model_name)

  • Updated health monitor API with engagement scoring and structured issue detection

Documentation & Configuration:

  • Complete rewrite of configuration documentation with all 16 new fields

  • New competitive positioning document comparing against mem0, Zep, Letta, Cognee, LangMem

  • Comprehensive feature documentation with architecture diagrams and code examples

  • Updated API reference with 15 new endpoints and new enum values

  • Updated Celery guide with new background task documentation

  • Updated README with v0.5.0 features and examples

Test Coverage:

  • 545 passing tests with updated assertions for auto-extracted entities

  • New comprehensive export/portability tests with compression support

  • Refactored integration tests for new API signatures

  • Added null safety and type conversion tests

  • Relaxed graph test assertions to account for dynamic entity extraction

Code Quality:

  • pyright: 0 errors

  • ruff: all checks passed

  • Backward compatible — all existing APIs remain functional

  • 242 files changed, +56,560 / -5,480 lines

Diagram Walkthrough: #26

Assets 2
Loading