MCP & Tool-Aufrufe

Metronix Memory

mtrnix/metronix-memory

Self-hosted memory infrastructure for AI agents featuring MCP-native integration, hybrid RAG, a temporal knowledge graph, and an ontology layer, designed for local-model friendliness and durable, agent-scoped context.

Offizielles Repository öffnen ↗ Projekt-Startseite ↗
★ 39Sterne
⑂ 7Forks
78Offene Issues
PythonSprache
Apache-2.0Lizenz
Q@project.QualityScoreRedaktionelle Bewertung

Projekt-Screenshots

Screenshot von Metronix Memory Screenshot von Metronix Memory

Übersicht

Metronix Memory provides a comprehensive backend that allows AI agents to ingest files and SaaS knowledge, retrieve context using dense, sparse, and graph retrieval, store durable facts and preferences, and maintain long-lived knowledge freshness. It contrasts itself with standalone vector databases by offering built-in ingestion, sparse retrieval, graph context, and operational APIs, and improves upon simple chat history by providing structured facts, temporal knowledge, and reusable context for any MCP-native agent. The architecture follows a strict one-way dependency model across six layers, from core configuration up to the API and channel integrations.

Wichtige Funktionen

  • MCP-native server endpoint
  • Hybrid RAG (Dense vectors + SPLADE sparse vectors + Neo4j graph context)
  • Temporal knowledge graph and ontology layer
  • Durable agent memory scoped by workspace and agent ID
  • Freshness checks and stale memory conflict resolution
  • Local-model friendly (bundled Ollama container for embeddings and graph extraction)
  • Native connector framework for Confluence, Jira, Notion, GitHub, Google Drive, Slack
  • OpenAI-compatible API
  • Metronix Admin Console for UI-based administration

Voraussetzungen, Installation und Schnellstart

Clone the repository: `git clone https://github.com/mtrnix/metronix-memory.git && cd metronix-memory`. Configure the environment: `cp .env.example .env` and generate an MCP API key: `printf '\nMETRONIX_MCP_API_KEY=%s\n' "$(openssl rand -hex 32)" >> .env`. Launch the stack: `docker compose up -d --build`. Verify the installation: `curl http://localhost:8000/health`. Alternatively, use the quick install script: `curl -fsSL https://mtrnix.com/install.sh | bash`.

Nutzung

After starting the backend, authenticate via the REST API to obtain a JWT token. Store a memory record by sending a POST request to `http://localhost:8000/api/v1/memory/records` with the content, agent_id, scope, and kind. Retrieve memories via a POST request to `http://localhost:8000/api/v1/memory/search`. For MCP usage, connect an MCP client (like Cursor or Claude Desktop) to `http://localhost:8000/mcp` and use tools like `metronix_memory_store` and `metronix_memory_search`.

Modellkompatibilität und Anwendungsfälle

Local-model friendly. The default Docker setup includes an Ollama container that pulls `nomic-embed-text` for embeddings and `qwen2.5:3b` for knowledge-graph extraction. External chat LLMs (like `deepseek-chat`) are only required if using the optional Open WebUI or if Metronix is configured to generate answers itself.

Lizenz- und Risikohinweise

Apache-2.0

Editorial verification 2026-08-02: repository URL, owner, description, license and repository statistics were reviewed. License metadata: Apache-2.0. README was fetched for the channel draft; re-check repository dependencies, releases and model terms before production use.

Veröffentlichung und Wartung

Not stated in the repository metadata

MemPalace

mempalace/mempalace

★ 58KPython

QwenPaw

agentscope-ai/qwenpaw

★ 33,9KPython

Composio

composiohq/composio

★ 29,5KTypeScript