Cursor
MCP · zero code changes
Persistent memory for AI coding: your assistant remembers project architecture, decisions and debugging context across every session.
Add to
.cursor/mcp.jsonIntegrations foryour framework
Your LangChain, CrewAI or LlamaIndex agents lose everything when a session ends: users re-explain preferences, bots repeat the same questions, multi-agent workflows restart from scratch. Install the Dakera integration and your agents recall context in under 50ms across sessions, with no embeddings running locally and no data leaving your server.
- 15integrations and plugins
- 14core MCP memory tools, 100+ via profiles
- 1self-contained Docker image to run
Pick your framework
Official packages on PyPI and npm, a config block for MCP clients, a Dify Marketplace plugin and a community plugin for Hermes Agent.
MCP clients
Connect Dakera to any MCP-compatible AI tool with zero code changes. 14 core memory tools are available instantly, and 100+ through profiles for power users.
Claude Desktop
MCP · zero code changes
Give Claude persistent cross-session memory: preferences, project context and knowledge that survive restarts.
Add to
claude_desktop_config.jsonWindsurf
MCP · zero code changes
Persistent memory for Cascade AI. It remembers project architecture, coding decisions and debugging context across every Windsurf session.
Add to
~/.codeium/windsurf/mcp_config.jsonPython
Drop-in memory and retrieval classes for the Python agent frameworks, installed with pip.
LlamaIndex
Python · llamaindex-dakera
Server-side embedding, so RAG needs no OpenAI API key.
DakeraMemoryStore- Agent memory
DakeraIndexStore- Replaces local vector indices
pip install llamaindex-dakeraStrands Agents
Python · strands-dakera
A dakera_memory tool plus a drop-in DakeraMemoryStore for the Strands agent loop, with decay-weighted recall ranked by importance × recency × relevance.
pip install strands-dakeraAgent Squad
Python · TypeScript · Swift
DakeraRetriever grounds agents in server-side semantic search over a Dakera namespace, with no local embedding model. Ships in three languages.
pip install "agent-squad[dakera]"JavaScript and TypeScript
Memory for LangChain.js chains and the Vercel AI SDK, installed with npm.
LangChain.js
TypeScript · @dakera-ai/langchain
DakeraMemory and DakeraVectorStore for LangChain.js chains. Full TypeScript types, compatible with Node.js ≥ 20.
npm install @dakera-ai/langchainVercel AI SDK
TypeScript · @dakera-ai/ai-sdk
Cross-session memory through the AI SDK's two extension points, language model middleware and tools. No model or provider changes required.
npm install @dakera-ai/ai-sdkPlatforms, governance and community
Dify adds memory to no-code and low-code agent platforms without an SDK integration. TealTiger keeps governance state for LLM applications (policy decisions, cost tracking and delegation audit trails) in Dakera's decay-weighted memory engine. Community integrations are built and maintained by third parties, not official Dakera packages: we link them so you can find them, so check each project's licence and support terms before relying on it.
Dify
Marketplace plugin · Dakera Memory
Six memory tools (store, recall, search, get, update, forget) for Dify Agent, Chatflow and Workflow apps. Install from the Marketplace, point it at your server, done.
Dify Marketplace → Dakera Memory
Hermes Agent
Python · community-maintained · by @Xyzjesus
Persistent memory for Nous Research's Hermes Agent: per-turn retention, pre-turn recall, and recall, store, search and forget tools, backed by your own Dakera server. Not an official Dakera package.
hermes plugins install Xyzjesus/hermes-dakeraHow integrations work
Every integration is a thin adapter between the framework's memory or vector-store interface and the Dakera REST API. No embeddings run locally: the Dakera server handles them with its built-in ONNX inference engine.
- Your frameworkLangChain, CrewAI, LlamaIndex, AutoGen and more, with the integration plugged in
- Dakera SDKPython, JS, Rust, Go or the REST API, over HTTPS
- Dakera serverEmbeds, indexes and retrieves with HNSW, BM25 and a knowledge graph
- Memory storedPersistent and searchable, AES-256-GCM encrypted at rest
| Feature | What Dakera provides |
|---|---|
| Embedding | On-device ONNX model on the server, with zero external API calls |
| Vector search | HNSW with IVF + SPFresh, BM25 hybrid reranking |
| Memory decay | Access-weighted importance, configurable half-life |
| Sessions | Per-session memory grouping and lifecycle management |
| Cross-agent network | Agents share knowledge via the cross-agent graph API |
Prerequisites
All integrations require a running Dakera server. No waitlist and no managed service required: run the binary on your own infrastructure, as a single self-contained Docker image with no external dependencies.
All framework integrations are published on PyPI and npm today. For persistent storage, see the Deployment guide, then pick your framework above and follow its integration docs.
docker run -d \
--name dakera \
-p 3000:3000 \
-e DAKERA_ROOT_API_KEY=dk-mykey \
ghcr.io/dakera-ai/dakera:latestIntegrated in under 10 minutes
One Docker command starts the server. One pip install adds the integration. Your LangChain or CrewAI agents start remembering, with no embedding service, no external database and no data leaving your infrastructure.
Further readingMulti-agent memory systemsMemory patterns libraryAPI reference