Add Dakera — Self-Hosted Memory Server for Production AI Agents

[ferhimedamine]

Dakera — Production AI Agent Memory Infrastructure

Repository: https://github.com/dakera-ai/dakera-deploy

Suggested category: Feature Store / Model Serving / Agent Infrastructure

What it does:
Dakera is a self-hosted memory server built in Rust that gives production AI agents persistent, decay-weighted vector memory across sessions. It solves the "amnesia problem" in deployed agent systems — where agents lose all conversation context between sessions and can't learn from past interactions.

Production-relevant features:

• Decay-weighted importance — memories degrade naturally over time (prevents unbounded index growth in production)
• Hybrid BM25 + vector search — sub-100ms retrieval combining keyword precision with semantic understanding
• Session isolation — namespace-based memory scoping for multi-tenant deployments
• Knowledge graph — automatic entity extraction and relationship mapping
• Multi-agent support — agents share memories with access control
• REST API + MCP — integrate with any LLM agent framework
• SDKs: Python, JavaScript, Rust, Go
• Self-hosted — Docker Compose deploy, full data sovereignty, no external dependencies

Why it's production-grade:

• Written in Rust for performance and memory safety
• RocksDB + HNSW index for durable, low-latency storage
• Benchmarked: 87% recall accuracy on LoCoMo 1540-question conversational memory benchmark
• ONNX inference for embeddings (no GPU required)
• Designed for always-on agent systems with hundreds of concurrent sessions