# Dakera > Dakera is a self-hosted AI agent memory server written in Rust. It gives AI agents persistent, decay-weighted memory with hybrid BM25+HNSW vector search, a knowledge graph, session management, and 14 core MCP tools (86+ available via profiles) — all on your own infrastructure. ## Docs - [Quick Start](https://dakera.ai/docs/quickstart): Install Dakera in under 5 minutes with Docker or Docker Compose - [MCP Server](https://dakera.ai/docs/mcp-server): Connect 14 core MCP tools (86+ available via profiles) to Claude Desktop, Claude Code, Cursor, or Windsurf - [REST API](https://dakera.ai/docs/api): Complete REST API reference — all endpoints for memory, recall, search, sessions - [Python SDK](https://dakera.ai/docs/python-sdk): dakera PyPI package reference — pip install dakera - [TypeScript SDK](https://dakera.ai/docs/typescript-sdk): dakera npm package — npm install @dakera-ai/dakera - [Rust SDK](https://dakera.ai/docs/rust-sdk): dakera-client crate — cargo add dakera-client - [Go SDK](https://dakera.ai/docs/go-sdk): dakera-go — go get github.com/dakera-ai/dakera-go@latest - [CLI](https://dakera.ai/docs/cli): dakera-cli — cargo install dakera-cli - [Deployment](https://dakera.ai/docs/deployment): Docker, Kubernetes (Helm), systemd, and cloud guides - [Architecture](https://dakera.ai/docs/architecture): Storage engine, retrieval pipeline, embedding layer, query classification - [Security](https://dakera.ai/docs/security): AES-256-GCM, scoped API keys, rate limiting, injection detection - [High Availability](https://dakera.ai/docs/high-availability): Gossip protocol, leader election, sharding, replication ## Integrations - [Integration Hub](https://dakera.ai/integrations/): All framework integrations — LangChain, CrewAI, LlamaIndex, AutoGen, LangChain.js, Cursor, Claude Desktop, Windsurf, TealTiger - [Cursor MCP](https://dakera.ai/integrations/cursor): Persistent memory for AI coding — 14 core MCP tools (86+ available via profiles), project context across sessions - [Claude Desktop MCP](https://dakera.ai/integrations/claude-desktop): Give Claude Desktop persistent cross-session memory — zero code changes - [Windsurf MCP](https://dakera.ai/integrations/windsurf): Give Windsurf Cascade AI persistent memory — project context, decisions, and patterns across sessions - [LangChain (Python)](https://dakera.ai/integrations/langchain): DakeraMemory + DakeraVectorStore for LangChain chains — pip install langchain-dakera - [CrewAI](https://dakera.ai/integrations/crewai): DakeraStorage as CrewAI long-term memory — pip install crewai-dakera - [LlamaIndex](https://dakera.ai/integrations/llamaindex): DakeraMemoryStore + DakeraIndexStore for LlamaIndex agents — pip install llamaindex-dakera - [AutoGen](https://dakera.ai/integrations/autogen): DakeraMemory for AutoGen agents and multi-agent teams — pip install autogen-dakera - [LangChain.js](https://dakera.ai/integrations/langchain-js): DakeraMemory + DakeraVectorStore for TypeScript chains — npm install @dakera-ai/langchain - [TealTiger](https://dakera.ai/integrations/tealtiger): Persistent governance state for LLM applications — DakeraCostStorage, DakeraDecisionStore, DakeraDelegationHelper ## Comparisons - [Dakera vs Mem0](https://dakera.ai/compare/mem0): Self-hosted vs managed API trade-off — both score high on LoCoMo - [Dakera vs Zep](https://dakera.ai/compare/zep): Closest competitor — both offer memory + knowledge graph - [Dakera vs Pinecone](https://dakera.ai/compare/pinecone): Agent memory engine vs managed vector database - [Dakera vs ChromaDB](https://dakera.ai/compare/chromadb): Production memory vs development embedding database - [Dakera vs Qdrant](https://dakera.ai/compare/qdrant): Both Rust, both fast — Dakera adds memory semantics - [Dakera vs Weaviate](https://dakera.ai/compare/weaviate): Agent memory vs multi-modal vector search - [Dakera vs Letta](https://dakera.ai/compare/letta): Deterministic retrieval vs LLM-powered memory management - [Dakera vs Cognee](https://dakera.ai/compare/cognee): Self-hosted Rust engine vs Python knowledge-graph-focused memory framework - [Dakera vs LangMem](https://dakera.ai/compare/langmem): Self-hosted Rust engine vs LangChain's memory abstractions - [Dakera vs Milvus](https://dakera.ai/compare/milvus): Agent memory engine vs open-source vector database built for scale - [Dakera vs pgvector](https://dakera.ai/compare/pgvector): Self-hosted memory engine vs PostgreSQL vector similarity extension - [Dakera vs Redis](https://dakera.ai/compare/redis): Self-hosted memory engine vs in-memory database with vector search module - [Mem0 Alternatives](https://dakera.ai/compare/mem0-alternatives): Top Mem0 alternatives in 2026 — Dakera, Zep, Letta, Cognee, LangMem and more, with benchmarks and trade-offs - [Mem0 Self-Hosting Comparison](https://dakera.ai/compare/mem0-self-hosting): Dakera (single Docker container, zero deps) vs Mem0 self-hosted (Qdrant + PostgreSQL + LLM service) ## Key Resources - [Benchmark Results](https://dakera.ai/benchmark/): 88.2% on LoCoMo 1,540Q — Cat1 86.9%, Cat2 85.4%, Cat3 73.9%, Cat4 91.0% - [Blog](https://dakera.ai/blog/): Technical articles on agent memory, MCP servers, and Rust infrastructure - [Changelog](https://dakera.ai/docs/): Version history for all Dakera artifacts ## Blog Posts - [Introducing Dakera](https://dakera.ai/blog/introducing-dakera): What Dakera is and why we built it - [How Agent Memory Works](https://dakera.ai/blog/how-agent-memory-works): Architecture of store, recall, and decay - [Dakera as an MCP Memory Server](https://dakera.ai/blog/dakera-mcp-memory-server): 14 core tools (86+ available via profiles) for Claude, Cursor, Windsurf - [Why Rust for AI Memory](https://dakera.ai/blog/why-rust-for-ai-memory): Performance, safety, and 44 MB binary - [Open Core Model](https://dakera.ai/blog/why-dakera-is-open-core): What is MIT-licensed and what is proprietary - [Benchmark Methodology](https://dakera.ai/blog/dakera-benchmark-methodology): How we measure recall quality with LoCoMo - [Best AI Agent Memory Frameworks 2026](https://dakera.ai/blog/best-ai-agent-memory-frameworks-2026): Dakera vs Mem0 vs Letta vs Zep vs Hindsight — benchmarks and architecture compared - [LangGraph Persistent Memory](https://dakera.ai/blog/langgraph-persistent-memory): Add semantic long-term memory to LangGraph agents — beyond MemorySaver, with cross-session recall, decay, and knowledge graphs - [MCP Memory Server Comparison 2026](https://dakera.ai/blog/mcp-memory-server-comparison-2026): Dakera vs Hindsight, MemPalace, Mem0 — feature matrix and benchmark results - [MCP Memory Server Setup Guide](https://dakera.ai/blog/mcp-memory-server-setup-guide): Add persistent memory to Claude Desktop, Claude Code, Cursor, and Windsurf in under 10 minutes - [Self-Hosted AI Memory Guide](https://dakera.ai/blog/self-hosted-ai-memory-complete-guide): Complete guide to deploying private AI agent memory — no cloud dependencies - [Multi-Agent Memory Systems](https://dakera.ai/blog/building-multi-agent-memory-systems): Architecture patterns for shared memory across multiple agents - [Vector DB vs Agent Memory](https://dakera.ai/blog/vector-database-vs-agent-memory): Why vector databases alone aren't enough for agent memory workloads - [Knowledge Graphs for Agents](https://dakera.ai/blog/knowledge-graphs-ai-agents): Entity extraction, relationship inference, and graph traversal - [Temporal Memory & Decay](https://dakera.ai/blog/temporal-memory-ai-agents): Importance scoring, decay strategies, and memory lifecycle - [Memory Decay Is Native, Not a Plugin](https://dakera.ai/blog/memory-decay-native-not-a-plugin): Why decay is a first-class feature in Dakera, not bolted on - [MCP Protocol Explained](https://dakera.ai/blog/mcp-protocol-ai-tools): How AI tools communicate with memory servers via Model Context Protocol - [Memory Benchmarks 2026](https://dakera.ai/blog/ai-agent-memory-benchmarks-2026): LoCoMo, evaluation methodology, and where the field is heading - [TealTiger Integration](https://dakera.ai/blog/dakera-tealtiger-integration): Persistent governance memory for LLM apps with TealTiger + Dakera - [Introducing the Dakera Playground](https://dakera.ai/blog/introducing-dakera-playground): Try agent memory in your browser — no install required - [Playground Now Live](https://dakera.ai/blog/playground-now-live): The Dakera Playground is live — interactive agent memory in the browser ## Memory Patterns - [Patterns Library](https://dakera.ai/patterns/): 22 production-ready agent memory patterns with code examples - [User Preference Recall](https://dakera.ai/patterns/user-preference-recall): Store and personalize based on user preferences - [Cross-Session Context](https://dakera.ai/patterns/cross-session-context): Maintain continuity across disconnected sessions - [Multi-Agent Shared Memory](https://dakera.ai/patterns/multi-agent-shared-memory): Multiple agents sharing and building on memories - [Hierarchical Memory](https://dakera.ai/patterns/hierarchical-memory): Short-term, long-term, and permanent memory tiers - [Knowledge Graph Entity Linking](https://dakera.ai/patterns/knowledge-graph-entity-linking): Extract entities and build traversable graphs - [Memory Compression](https://dakera.ai/patterns/memory-compression): Consolidate verbose memories into summaries - [Goal Tracking](https://dakera.ai/patterns/goal-tracking): Track user goals across conversations - [Feedback Loop Memory](https://dakera.ai/patterns/feedback-loop-memory): Learn from corrections to avoid repeating mistakes - [Context Window Management](https://dakera.ai/patterns/context-window-management): Token-aware memory injection for LLMs - [Batch Ingestion](https://dakera.ai/patterns/batch-ingestion): Bootstrap memory from existing data sources - [Conflict Resolution](https://dakera.ai/patterns/conflict-resolution): Resolve contradictory memories using recency, confidence, and explicit user confirmation - [Conversation Summarization & Decay](https://dakera.ai/patterns/conversation-summarization-decay): Consolidate conversations into persistent summaries while raw transcripts decay - [Emotional Context Tracking](https://dakera.ai/patterns/emotional-context-tracking): Store user emotional state signals for tone-aware, empathetic AI responses - [Episodic Replay](https://dakera.ai/patterns/episodic-replay): Reconstruct conversation history to understand context evolution across sessions - [Importance-Based Forgetting](https://dakera.ai/patterns/importance-based-forgetting): Score memories by importance and let low-value ones decay to bound memory growth - [Memory Permissions](https://dakera.ai/patterns/memory-permissions): Role-based access control for memory namespaces with scoped API keys - [Namespace Isolation](https://dakera.ai/patterns/namespace-isolation): Isolate memory by user, tenant, or agent for secure multi-tenant AI applications - [RAG-Augmented Memory](https://dakera.ai/patterns/rag-augmented-memory): Combine persistent agent memory with retrieved document chunks for richer AI responses - [Semantic Deduplication](https://dakera.ai/patterns/semantic-deduplication): Detect and merge semantically similar memories to keep memory stores clean and efficient - [Session Handoff](https://dakera.ai/patterns/session-handoff): Transfer conversation context between agents or sessions for bot-to-human escalation - [Temporal Event Tracking](https://dakera.ai/patterns/temporal-event-tracking): Track events over time and answer queries like "what changed last week" with time-aware recall - [Tool Usage Learning](https://dakera.ai/patterns/tool-usage-learning): Store tool usage outcomes so agents learn which tools work for which tasks ## Optional - [GitHub org](https://github.com/dakera-ai): Source code for SDKs, CLI, MCP server, and deploy tooling - [Docker image](https://github.com/dakera-ai/dakera/pkgs/container/dakera): ghcr.io/dakera-ai/dakera:latest ## Artifacts (server: v0.11.107, py: v0.12.12, js/rs: v0.11.106, go: v0.11.103) | Artifact | Version | Install | |----------|---------|---------| | dakera (server) | 0.11.107 | docker pull ghcr.io/dakera-ai/dakera:latest | | dakera-py | 0.12.12 | pip install dakera | | dakera (npm) | 0.11.106 | npm install @dakera-ai/dakera | | dakera-client (Rust) | 0.11.106 | cargo add dakera-client | | dakera-go | 0.11.103 | go get github.com/dakera-ai/dakera-go@latest | | dakera-cli | 0.7.1 | cargo install dakera-cli | | dakera-mcp | 0.10.12 | binary download from GitHub releases |