AI Agents Forget.
HippocampAI Remembers.
Production-ready memory layer with knowledge graphs, hybrid retrieval, prospective memory, multi-agent collaboration, and 120+ API methods. Give your AI systems human-like memory capabilities.
Works with
Get started in minutes
Install as a lightweight library or spin up the full platform with Docker
Install
Get started with pip
pip install
hippocampaiInitialize
Create a memory client
from hippocampai
import MemoryClient
client = MemoryClient()Remember
Store and recall
client.remember(
"Your first memory",
user_id="alice"
)Install only what you need
How HippocampAI Works
A robust pipeline designed for reliability and performance at scale
Memory Types
- Preferences
- Facts & knowledge
- Events & conversations
- Procedural (behavioral)
- Prospective (future)
- Custom metadata
Processing
- Multi-provider embeddings
- Knowledge graph extraction
- Importance scoring
- Relevance feedback loops
- Context assembly + token budgeting
Storage & Caching
- Qdrant vector DB (HNSW)
- Redis caching (50-100x)
- DuckDB (sessions + consolidation)
- Tiered storage (hot/warm/cold)
- Export/import (JSON, Parquet, CSV)
Sleep Phase
- Memory consolidation
- Importance decay
- Auto-healing
- Predictive analytics
- Session summarization
Everything you need for AI memory
From knowledge graphs to multi-agent collaboration — all the primitives for building memory-enabled AI applications
Knowledge Graph
Real-time entity and relationship extraction on every remember() call. Build structured knowledge automatically from unstructured text.
3-way RRF fusion: vector + BM25 + graph retrieval
Hybrid Retrieval
Combines semantic search, BM25 keyword matching, and graph-aware retrieval with Reciprocal Rank Fusion for optimal results.
40% accuracy improvement over vector-only search
Multi-Agent Collaboration
Shared memory spaces for agent coordination. Multiple AI agents can read, write, and reason over collective memories.
Agent-scoped access with shared namespaces
Sleep Phase Consolidation
Bio-inspired memory consolidation merges related memories, decays importance scores, and prunes low-value data automatically.
Inspired by human memory consolidation during sleep
Predictive Analytics
Pattern detection, habit tracking, and behavioral insights. Predict memory usage patterns and forecast future needs.
Cross-session analytics with temporal reasoning
Auto-Healing
Automatic detection and repair of memory issues. Health monitoring, duplicate detection, and quality tracking built in.
Self-monitoring with configurable health thresholds
Memory Triggers
Event-driven actions via webhooks, websocket, and log triggers. React to memory events in real-time across your system.
Configurable trigger rules with filter conditions
Procedural Memory
Self-optimizing prompts via learned behavioral rules. Your AI learns how to communicate better over time.
Feedback loops with exponential decay scoring
Plugin System
Extensible architecture with custom processors, scorers, retrievers, and filters. Build exactly the memory system you need.
Custom schemas, tiered storage, export/import
SaaS Platform
Full multi-tenant platform with authentication, rate limiting, Celery background tasks, and a React dashboard.
pip install hippocampai[saas] for full platform
Importance Scoring
Not all memories are equal. Automatic importance scoring with access-based boosting and configurable decay rates.
Time-weighted scoring with relevance feedback
Memory Namespaces
Hierarchical organization with permissions. Isolate memory spaces per user, session, team, or custom scope.
Bi-temporal facts with time-travel queries
Prospective Memory
"Remembering to remember" — schedule future actions with time-based or event-based triggers. Supports daily, weekly, monthly, and custom cron recurrence.
POST /v1/prospective/ · 9 endpoints · croniter recurrence
Batch Operations
Store, fetch, or delete hundreds of memories in a single API call. Deduplication endpoint with dry_run mode. Individual failures never abort the batch.
POST /v1/memories/batch · /batch/get · /batch/delete · /deduplicate
Session Management
Track multi-turn conversations as structured sessions. Automatic LLM-powered summarization when sessions hit a threshold. Semantic search across sessions.
create / update / complete / delete · session summarization
Context Assembly
Smart context-pack generation for RAG. Token budget management automatically selects and ranks memories that fit within your model's context window.
token-budget aware · automatic context pack generation
Bi-Temporal Facts
Track both valid time (when a fact was true) and transaction time (when recorded). Run time-travel queries to reconstruct any past state.
valid_time + transaction_time · historical state reconstruction
Remote Backend
Point the Python library at a running HippocampAI API server instead of embedding storage locally. Great for microservices and serverless.
RemoteBackend(url=...) · 90s timeout · all SDK methods
Simple API, powerful features
Get started in minutes with our intuitive Python SDK. 120+ methods covering memory storage, retrieval, batch operations, prospective memory, knowledge graphs, multi-agent coordination, and more.
pip install hippocampaiPyPIfrom hippocampai import MemoryClient
# Initialize client
client = MemoryClient()
# Store a memory
client.remember(
"User prefers oat milk and works remotely on Tuesdays",
user_id="alice",
type="preference"
)
# Recall relevant memories
results = client.recall("work preferences", user_id="alice")
for r in results:
print(f"{r.memory.text} (relevance: {r.score:.2f}")Works with your stack
Seamlessly integrate with popular LLM providers, storage backends, and AI frameworks
LLM & Embedding Providers
Storage Backends
Frameworks & Agent Platforms
How we compare
See how HippocampAI stacks up against other memory solutions
| Feature | HippocampAIOpen Source | Mem0Cloud | LangChainMemory | CustomSolution |
|---|---|---|---|---|
| Semantic Search | ||||
| Knowledge Graphonly us | ||||
| Graph-Aware Retrieval (3-way RRF)only us | ||||
| Auto-Deduplication (+ API endpoint)only us | ||||
| Hybrid Retrieval (BM25 + Vector) | ||||
| Multi-Agent Collaboration | ||||
| Session Management | ||||
| Prospective Memory (future triggers)only us | ||||
| Batch Operations (store/get/delete)only us | ||||
| Context Assembly + Token Budgetingonly us | ||||
| Bi-Temporal Facts (time-travel queries)only us | ||||
| Memory Triggers & Webhooks | ||||
| Procedural Memory | ||||
| Sleep Phase Consolidationonly us | ||||
| Predictive Analytics | ||||
| Auto-Healing | ||||
| Export / Import (JSON, Parquet, CSV) | ||||
| Offline Mode + Operation Queueonly us | ||||
| Self-Hosted Option | ||||
| Built-in Dashboard UI | ||||
| Plugin System | ||||
| Open Source (Apache 2.0) |
Built for every use case
HippocampAI powers memory for diverse AI applications across industries
Conversational AI
Build chatbots that remember user preferences, conversation history, and context across sessions.
AI Agents
Create autonomous agents that learn from interactions and make informed decisions based on past experiences.
Document Q&A
Enable semantic search over large document collections with context-aware retrieval.
Multi-User Systems
Manage separate memory spaces for different users while maintaining privacy and isolation.
Personalization
Deliver tailored content and suggestions by learning from user behavior and preferences.
Performance that scales
Built for production workloads — run the benchmark suite yourself to verify
Latency Benchmarks
Measured on commodity hardware · Qdrant HNSW M=32 EF=128
Storage Efficiency
Open source & community driven
Join the community building the future of AI memory
Frequently asked questions
Everything you need to know about HippocampAI
HippocampAI is an open-source, production-ready memory engine that gives AI systems human-like memory capabilities. Named after the brain region responsible for memory formation, it provides persistent memory storage, knowledge graphs, hybrid retrieval, prospective memory, batch operations, multi-agent collaboration, and 120+ API methods for building memory-enabled AI applications.
Still have questions?
Ask on GitHub DiscussionsChangelog & Roadmap
Stay up to date with the latest features and improvements
Recent Releases
- Batch memory endpoints: store/get/delete N memories in one request
- Deduplication API: POST /v1/memories/deduplicate with dry_run support
- Single-memory GET: GET /v1/memories/{memory_id} with 404 on miss
- Prometheus /metrics endpoint in main app with graceful fallback
- RemoteBackend URL patterns corrected; timeout raised to 90s
- QueryRouter stem-prefix matching — plurals and conjugations now work
- Groq retry capped at 3 × 20s (was 5 × 60s, could hang ~5 min)
- Prospective Memory — time-based and event-based future triggers
- ProspectiveMemoryManager with full pending→triggered→completed lifecycle
- Recurrence: daily, weekly, monthly, custom_cron via croniter
- Embedding-similarity event matching for trigger conditions
- 9 new REST endpoints under /v1/prospective/ (CRUD, parse, evaluate)
- Prospective intents surface automatically during recall()
- Background evaluation loop in BackgroundTaskManager
- Knowledge graph with real-time entity extraction
- Graph-aware retrieval (vector + BM25 + graph RRF)
- Relevance feedback loop with decay scoring
- Memory triggers (webhooks, websocket, log actions)
- Procedural memory & prompt self-optimization
- Embedding model migration with Celery
- Multi-agent collaboration with shared memory
- Predictive analytics & pattern forecasting
- Auto-healing memory system
- React dashboard with full analytics UI
- Plugin system (processors, scorers, retrievers)
- Memory namespaces with permissions
- SaaS platform (auth, rate limiting, Celery)
- Export/import (JSON, Parquet, CSV)
- Tiered storage (hot/warm/cold)
- Offline mode with operation queueing
- Bi-temporal facts with time-travel queries
- Context assembly with token budgeting
Roadmap
Ready to get started?
HippocampAI is open source and free to use. Star us on GitHub to show your support and stay updated with new releases.
Get notified about new releases