Skip to content

Suggestion: Add Dakera — Memory Infrastructure for AI Agent Operations #185

Description

@ferhimedamine

Dakera — Self-Hosted AI Agent Memory Server

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

Suggested section: Data Management / Feature Store / Model Serving Infrastructure

Description:
Dakera is a self-hosted memory server for AI agent systems. As LLM agents move into production (customer support, code assistants, autonomous workflows), they need persistent memory that survives across sessions. Dakera provides this as infrastructure — a Rust-based server with decay-weighted vector memory, hybrid search, and knowledge graph capabilities.

Key capabilities:

  • Decay-weighted importance scoring (production-safe: prevents unbounded memory growth)
  • Hybrid BM25 + vector retrieval (sub-100ms query latency)
  • Session-based memory management for multi-tenant agent deployments
  • Knowledge graph with automatic entity extraction
  • REST API + MCP (Model Context Protocol) integration
  • SDKs in Python, JavaScript, Rust, Go
  • Docker Compose deployment, self-hosted

Relevance to MLOps:
As AI agents become operational (not just experimental), they need memory infrastructure the same way ML models need feature stores and model registries. Dakera is that infrastructure layer — designed for always-on, multi-agent production systems.

Built in Rust. Benchmarked at 87% recall on LoCoMo 1540Q conversational memory.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions