Changelog

All notable changes to HippocampAI, newest first. Full history on GitHub Releases

v0.5.1Latest ReleaseApril 9, 2026

Batch memory operations, deduplication API, RemoteBackend fixes, and QueryRouter improvements.

Added

Batch memory endpoints — POST /v1/memories/batch, /batch/get, /batch/delete
Deduplication endpoint — POST /v1/memories/deduplicate with dry_run=true support
Single-memory GET — GET /v1/memories/{memory_id}, returns 404 if not found
Prometheus /metrics endpoint in main app.py with 501 fallback when prometheus-client not installed
SQL fallback in consolidation/db.py — creates tables inline when .sql file is absent
Test suite for remote backend — 9 unit tests covering URL construction and consolidation DB

Changed

RemoteBackend URL patterns corrected — all 6 methods had colon-style paths, now match actual FastAPI routes
RemoteBackend default timeout raised from 30s to 90s
RemoteBackend filter merging — session_id, min_importance, after, before now merged into filters dict
QueryRouter matching switched to stem-prefix — plurals/conjugations now work (e.g. 'habits' → 'habit')
Groq retry config reduced from 5×60s to 3×20s with explicit 30s HTTP timeout
pyproject.toml: added **/*.sql glob so SQL files are included in installed package

Fixed

/api/consolidation/status returning HTTP 500 due to missing SQL file at install time
RemoteBackend.recall() silently returning empty results when server URL used colon-style verb paths
recall() returning 0 results for plural or conjugated queries against QueryRouter-routed endpoints
Groq LLM calls hanging for up to 5 minutes during transient API errors
npm ci failing during Docker frontend build due to missing package.json

v0.5.1Prospective MemoryFebruary 13, 2026

First-class Prospective Memory — a new cognitive capability for AI agents to remember to do things in the future.

Added

PROSPECTIVE memory type added to MemoryType enum
Two trigger modes: time-based (datetime/cron) and event-based (keywords, regex, embedding similarity)
ProspectiveIntent model with full lifecycle: pending → triggered → completed → expired → cancelled
ProspectiveMemoryManager with CRUD, context evaluation, NL parsing, recurrence, expiry, consolidation
Recurrence support: none, daily, weekly, monthly, custom_cron (via croniter)
Priority-based ordering (0–10) for triggered intents
Recall pipeline integration: triggered intents surface automatically as RetrievalResult objects
ON_PROSPECTIVE_TRIGGER event added to trigger system
Background evaluation loop in BackgroundTaskManager
2 new Celery tasks: evaluate_prospective_triggers, expire_prospective_intents
9 new REST API endpoints under /v1/prospective/ (intents CRUD, parse, evaluate, consolidate, expire)
4 new config fields: enable_prospective_memory, prospective_max_intents_per_user, prospective_eval_interval_seconds, half_life_prospective
Library exports: ProspectiveIntent, ProspectiveMemoryManager, ProspectiveStatus, ProspectiveTriggerType, RecurrencePattern
33 unit tests covering CRUD, trigger matching, recurrence, lifecycle, NL parsing

v0.5.0Major ReleaseFebruary 11, 2026

Intelligent Memory Features — knowledge graph, feedback loops, triggers, procedural memory, and embedding migration.

Added

Memory Relevance Feedback Loop with 30-day half-life EMA scoring
Memory Triggers — on_remember, on_recall, on_update, on_delete, on_conflict, on_expire events
Graph-Aware Retrieval — 3-way RRF fusion (vector + BM25 + graph) via GraphRetriever
Procedural Memory / Prompt Self-Optimization with LLM rule extraction
Real-Time Incremental Knowledge Graph with JSON persistence
Embedding Model Migration with Celery background processing
GRAPH_HYBRID search mode added to SearchMode enum
PROCEDURAL memory type added to MemoryType enum
16 new config fields, 15 new REST API endpoints, 2 new Celery tasks

Changed

Scoring weights include graph and feedback components with auto-normalization via get_weights()
Decay half-lives include procedural type via get_half_lives()
fuse_scores() extended with graph and feedback kwargs

v0.4.0Major ReleaseJanuary 26, 2026

Memory Consolidation, Multi-Agent Collaboration, SaaS Platform, and React Dashboard.

Added

Memory Compaction System with DuckDB persistence and sleep-phase architecture
Bi-Temporal Facts with valid/transaction time and time-travel queries
Multi-Agent Shared Memory Spaces with granular permission controls
ClassifierService — consolidated agentic classifier with LLM support
Context Assembly System with token budget management
Custom Schema Support — define memory types without code changes
Comprehensive Audit Logging with retention policies
Usage Tracking and Analytics with per-user statistics
LangChain and LlamaIndex integration adapters
Plugin System with extensible processors, scorers, retrievers
Predictive Analytics Pipeline for pattern forecasting
Auto-Healing Pipeline for automatic memory repair
Tiered Storage (hot/warm/cold) for cost optimization
Offline Client with operation queueing and auto-sync
Namespace Management with hierarchical organization
Memory Portability — import/export in JSON, Parquet, CSV
WebSocket support for real-time updates
Complete React + TypeScript + Tailwind CSS Dashboard
10 new API route modules including bitemporal, collaboration, compaction, dashboard

Changed

Fixed

v0.3.0Stable ReleaseDecember 2025

Simplified API, documentation reorganization, and Docker deployment fixes.

Added

Fixed