The memory engine for AI agents.
Persistent, searchable, decay-weighted agent memory — built in Rust — as a single self-hosted binary.
dakera.ai · Documentation · Request Early Access →
ذاكرة — Dhākira — Arabic for memory
Every AI agent session starts from zero. Thousands of interactions — zero retained knowledge. You're paying to re-teach your agents the same things, every time.
Dakera gives your agents persistent, compounding memory backed by production-grade vector search, hybrid retrieval, knowledge graphs, and built-in ML embeddings — in one self-hosted Rust binary.
Stop managing five services. Deploy one binary.
1. Store → Your agent writes a memory (content + importance score)
2. Recall → Query by meaning — hybrid vector + BM25 retrieval returns ranked results
3. Decay → Memories auto-decay by access pattern; important ones rise, stale ones fade
Everything runs inside the same process — no sidecars, no embedding APIs, no message queues.
| Metric | Value |
|---|---|
| LoCoMo recall benchmark | 87.6% |
| p99 query latency | < 10 ms |
| Insert throughput | 27.4M / second |
| Binary size | ~44 MB |
| External runtime dependencies | 0 |
Benchmarked against the full 1,540-question LoCoMo conversational recall suite.
| Instead of running separately | Dakera provides |
|---|---|
| Qdrant / Pinecone / Weaviate | HNSW · IVF · SPFresh vector index |
| Elasticsearch / OpenSearch | BM25 full-text search engine |
| OpenAI / Cohere embeddings | On-device ONNX — zero API calls |
| Redis / Postgres memory layer | Decay-weighted agent memory with sessions |
| Neo4j | Built-in knowledge graph with cross-agent network |
# Pull and run
docker run -d -p 3300:3300 -e DAKERA_API_KEY=my-key ghcr.io/dakera-ai/dakera:latest
# Verify
curl http://localhost:3300/healthPython:
pip install dakerafrom dakera import DakeraClient
client = DakeraClient(base_url="http://localhost:3300", api_key="my-key")
# Store a memory
client.memories.store(
agent_id="my-agent",
content="User prefers TypeScript over Python",
importance=0.8,
tags=["preference"]
)
# Recall relevant memories
memories = client.memories.recall(agent_id="my-agent", query="language preferences")TypeScript:
npm install dakeraimport { DakeraClient } from 'dakera';
const client = new DakeraClient({ baseUrl: 'http://localhost:3300', apiKey: 'my-key' });
await client.memories.store({
agentId: 'my-agent',
content: 'User prefers TypeScript over Python',
importance: 0.8,
});
const memories = await client.memories.recall({
agentId: 'my-agent',
query: 'language preferences',
});Add Dakera to Claude, Cursor, or Windsurf in 30 seconds:
Claude Desktop — ~/Library/Application Support/Claude/claude_desktop_config.json:
{
"mcpServers": {
"dakera": {
"command": "dakera-mcp",
"env": { "DAKERA_API_URL": "http://localhost:3300", "DAKERA_API_KEY": "your-key" }
}
}
}Claude Code — .claude/settings.json in your project:
{
"mcpServers": {
"dakera": {
"command": "dakera-mcp",
"env": { "DAKERA_API_URL": "http://localhost:3300", "DAKERA_API_KEY": "your-key" }
}
}
}84 tools across: Memory CRUD · Vector Operations · Knowledge Graph · Sessions · Namespaces · Decay Engine · AutoPilot · Full-text Index
| Package | Version | Install |
|---|---|---|
| dakera-py |  |
pip install dakera |
| dakera-js |  |
npm install dakera |
| dakera-rs |  |
cargo add dakera-client |
| dakera-go |  |
go get github.com/dakera-ai/dakera-go |
| dakera-cli |  |
cargo install dakera-cli |
| dakera-mcp |  |
bundled with server |
| Package | Version | Install |
|---|---|---|
| dakera-langchain |  |
pip install langchain-dakera |
| dakera-llamaindex |  |
pip install llamaindex-dakera |
| dakera-crewai |  |
pip install crewai-dakera |
| dakera-autogen |  |
pip install autogen-dakera |
| dakera-langchain-js |  |
npm install langchain-dakera |
All SDKs and integrations are MIT licensed. The core engine and dashboard are proprietary.
# Docker (quickest)
docker run -d -p 3300:3300 -p 3500:3500 \
-e DAKERA_API_KEY=my-key \
ghcr.io/dakera-ai/dakera:latest
# Helm
helm install dakera oci://ghcr.io/dakera-ai/dakera-helm/dakera \
--namespace dakera --create-namespace \
--set dakera.rootApiKey=my-key \
--set minio.rootPassword=my-password→ dakera.ai/docs — Getting started, MCP setup, SDK references, configuration, deployment, architecture