Ingest from 300+ apps. Enrich with 42 AI agents. Query with 5 search modes. Self-hosted and free.
Run locally in one command. No account needed.
Why mem-dog?
Every time you ask an AI a question, your data travels to a distant data center. mem-dog ends that trade-off.
Data never leaves your network. Self-hosted with local-first models via Ollama — runs fully offline. No telemetry, no phone-home. Apache 2.0 licensed.
A 6-layer classification pipeline matches each piece of data to the smallest model that can handle it. Simple text gets a 4B model. Complex docs get a 27B one. Most queries finish in milliseconds.
Cloud AI bills grow linearly with usage. With local Ollama models, the marginal cost of a query is the electricity to run it. No per-seat licensing. No surprise invoices.
42 agents work in concert — classifying, extracting entities, generating embeddings, building knowledge graphs, and creating AI-powered viewpoints — all locally. Gets smarter the more you use it.
mem-dog = mem0 (memory) + Zep (knowledge graph) + Nango (integrations) + Dify (AI workflows) — in one platform.
How it Works
Into a private, queryable data layer — all automatic, all configurable.
300+ apps
Normalize & route
42 agents
10 memory types
Semantic RAG
15+ Channels
Features
Everything you need to build a private AI memory platform for your data.
Connect Slack, Gmail, GitHub, WhatsApp, Telegram, Discord, HubSpot, Salesforce, and 300+ more via Nango. OAuth2 with automatic token refresh and AES-256-GCM credential encryption.
Every piece of data is automatically classified, analyzed, summarized, and embedded by specialized agents. 60+ data types: PDFs, images, video, audio, code, medical records, geospatial data, IoT telemetry.
Vector (pgvector cosine), full-text (BM25), hybrid (vector + BM25 via RRF), graph (Graphiti BFS on Neo4j), and full (all signals merged). Plus 4 reranking strategies.
Dual-layer: Postgres + Graphiti/Neo4j. Facts have valid_at/invalid_at timestamps for point-in-time retrieval. "Who was CEO of Acme in 2024?" just works.
Conversational answers grounded in your data. Every claim is backed by numbered [1][2] citations linking back to source documents. Works with all 5 search modes.
Timeline, session, conversation, user, organizational, factual, episodic, semantic, custom, tracing. Default TTLs per type, ACLs (private/shared/public/restricted).
5 model tiers: small (4b) to omni. Per-agent-type assignment with fallback chains (Ollama local → Ollama Cloud → Gemini). Model Garden supports 10+ AI providers.
Each user gets unique webhook endpoints (whk_<ulid>) with HMAC-SHA256 verification, event logging, and stats. 25+ channel adapters normalize payloads into a universal format.
Python, TypeScript, Go, Rust, Ruby. ~120 methods each, plus a 7-method simple facade. Agent adapters for LangChain, CrewAI, and OpenAI function calling. MCP server for Claude Desktop.
Run on Docker Compose (local dev), GKE, or any Kubernetes cluster. No vendor lock-in, no usage fees. Apache 2.0 license. Your data never leaves your infrastructure.
System Architecture
Real-time processing, tenant isolation, and complete observability. Run locally or on Google Cloud.
| Component | Stack | Deployment |
|---|---|---|
| API | Python 3.12, FastAPI | GKE |
| UI | Next.js 14, TypeScript | Cloud Run |
| MCP Server | Python 3.12, FastMCP, SSE | GKE |
| Webhook Pipeline | Python 3.12, NATS, 42 agents | GKE |
| Webhook Gateway | Python 3.12, FastAPI, LiteLLM | GKE |
| Neo4j | Neo4j 5.26 Community + Graphiti | GKE |
| Nango | OAuth, token refresh, encryption | GKE |
| SDKs | Python, TypeScript, Go, Rust, Ruby | npm/pip/cargo |
Search Deep-Dive
Not just vector search. Every signal, every strategy, every query type — covered.
from mem_dog_client import MemDog
m = MemDog("http://localhost:8080")
# Hybrid search with MMR reranking
results = m.search(
"project timeline decisions",
mode="hybrid",
reranker="mmr",
limit=10
)
# RAG Chat with citations
answer = m.chat(
"What decisions were made?",
search_mode="hybrid",
reranker="cross-encoder"
)
print(answer.text)
# [1] Decision about roadmap...
# [2] Q3 planning notes...Comparison
Each tool does one thing well. mem-dog combines all of them — integrations, AI processing, memory, search, and a knowledge graph — into a single self-hosted system.
← Scroll to compare →
| Feature | Mem-Dog | Dify.ai | Mem0 | Zep | LangMem |
|---|---|---|---|---|---|
| Core Architecture | |||||
| Runs fully on-premise / at home | |||||
| Self-hosted AI models (Ollama) | |||||
| Near-zero marginal cost per query | |||||
| Data never leaves your network | |||||
| Single docker compose up | |||||
| Data & Integrations | |||||
| 300+ app integrations (Nango) | |||||
| Per-user webhook endpoints | |||||
| Multi-channel ingest (WhatsApp, Slack) | |||||
| File/image/video/audio ingest | |||||
| Versioned storage with full history | |||||
| AI & Intelligence | |||||
| 42 specialized AI agents | |||||
| 5 model tiers with smart routing | |||||
| Knowledge graph (Neo4j/Graphiti) | |||||
| Temporal fact tracking | |||||
| RAG chat with inline citations | |||||
| Memory & Search | |||||
| 10 memory types | |||||
| 5 search modes | |||||
| Memory compression (LLM summarization) | |||||
| Per-user scoping & ACLs | |||||
| Developer Experience | |||||
| Python SDK with LangChain/CrewAI adapters | |||||
| Multi-language SDKs (TS, Go, Rust, Ruby) | |||||
| Built-in conversational agent (DigiMe) | |||||
| Interactive playground in UI | |||||
mem-dog = Nango connects your apps. Dify builds AI workflows. Mem0 adds memory. Zep tracks facts. mem-dog does all of it — on hardware you own, at a fraction of the cost.
Use Cases
From personal knowledge management to enterprise compliance — one platform, infinite possibilities.
Capture from WhatsApp, email, Slack. Semantic search across everything. Never lose a conversation or idea.
Shared org memory across channels. Auto-classify meetings, decisions, action items. Queryable by anyone.
Ingest support tickets, CRM, chat logs. AI extracts sentiment and trends. 300+ app connections via Nango.
Ingest PDFs, papers, web pages, datasets. AI viewpoints and summaries. Semantic connections across sources.
Every mutation versioned. OpenTelemetry tracing. Per-item ACLs. Immutable audit trail.
GPS, biometric, weather, industrial sensors. Specialized agents for time-series and geospatial data.
Ingest contracts, NDAs, legal briefs. AI extracts clauses, obligations, deadlines. Temporal graph tracks amendments.
Process medical records, imaging reports, lab results. DICOM-aware agents. Knowledge graph links patients, conditions, treatments.
Ingest lectures, textbooks, course materials. AI generates study guides and flashcards. Students query their learning history.
Connect Salesforce, HubSpot, email. AI summarizes deal history, extracts action items. Search across all customer touchpoints.
RSS feeds, social media, news APIs. Real-time sentiment analysis. Track brand mentions across channels with temporal trends.
Zoom, Teams, Google Meet recordings. AI transcribes, extracts decisions, action items. Searchable meeting memory across your org.
Developer
SDKs in 5 languages, agent framework adapters, and a full REST API. Get started in minutes.
from mem_dog_client import MemDog
m = MemDog("http://localhost:8080")
# Add data
m.add("Project kickoff notes...", tags=["projects"])
# Semantic search
results = m.search("project timeline", mode="hybrid")
# RAG chat
answer = m.chat("What are the key milestones?")
# Knowledge graph
entities = m.entities(query="people involved")
related = m.related(entity_id="ent_01ABC...")Also available in Go, Rust, and Ruby. Full API docs at /docs.
FAQ
Get running in under 2 minutes. No account needed.
Enterprise inquiries: [email protected]