utekeOne Memory. Every Agent. Zero Cloud.

Give your AI a memory that never leaves your machine. Works with Claude, Cursor, Copilot, and any MCP-compatible agent.

🧠

Semantic Memory

AI remembers by meaning, not keywords. Local ONNX embeddings (768d) with usearch persistent HNSW index. Hybrid search via FTS5 + RRF.

📦

Single Binary

Zero dependencies. No Docker, no database server, no Python, no API keys.

🏷️

Tags + Namespaces + Metadata

Multi-agent isolation built-in. Entity, category, and key:value metadata enrichment. Tag management with list/rename/delete.

🕰️

Memory Aging

Hot/Warm/Cold tier tracking. Auto-cleanup stale memories.

🐚

Shell Hooks

Auto-load project context. bash/zsh/fish support.

🔒

Fully Offline

Local ONNX embeddings. No telemetry, no cloud, no API calls.

🏠

Rooms

Group memories by context (meetings, projects). Author attribution, semantic room recall, structured document generation.

⏳

Time-Travel Queries

Recall memories as they existed at any point in time. Temporal validity with valid_from/valid_until.

🔗

Relationship Graph

Link memories with typed edges. Supersedes, contradicts, references. BFS traversal with --related --depth.

📉

Smart Decay

Composite importance scoring. Pin critical memories. Auto-recalculate on access patterns.

🔌

MCP Server

Model Context Protocol server for AI tool integration. Expose memories to Claude, GPT, and other MCP clients.

📊

Benchmarks

Built-in uteke bench for performance testing. LongMemEval retrieval harness for accuracy evaluation.

Compare

Feature	uteke	Mem0	Letta	Cognee
Install	1 binary	pip + Docker + Qdrant	pip + Docker + Postgres	pip + Docker + Neo4j
API Keys	✅ None	❌ OpenAI/LLM key	❌ LLM key	❌ LLM + vector DB
Offline	✅ Fully	❌ Cloud embedding	❌ Needs server	❌ Needs LLM + DB
Recall Speed	~30ms	Network RTT	Network RTT	Network RTT
Privacy	✅ Local	⚠️ Sent to LLM	⚠️ Sent to LLM	⚠️ Sent to LLM

See full comparison →