IronRAG: memory tool for Hermes Agent

Self-hosted knowledge graph and RAG system with a 23-tool MCP server for agents, including Hermes

What IronRAG does

IronRAG is a self-hosted knowledge memory system for AI agents and teams. Documents are decomposed into entities, typed relationships and chunk-level evidence, and retrieval combines vector, lexical, graph-traversal and technical-fact lanes. Answers return citations to the underlying chunks. A native MCP server exposes 23 tools across documents, graph, web ingest and grounded ask, scoped per IAM token, and the README lists Hermes among compatible clients along with Claude Desktop, Claude Code, Cursor, Codex and OpenClaw.

The runtime is a Docker Compose stack with PostgreSQL and pgvector, Redis, a backend, a worker and a frontend, with a Helm chart for Kubernetes. Eight LLM providers ship in the catalog, including OpenAI, DeepSeek, Qwen, OpenRouter, MiniMax and Ollama, and every request is recorded in a USD cost catalog. Ingest is code-aware with tree-sitter parsing for 15 languages, document extraction runs on CPU through Docling, long jobs resume after restarts, and backups are tar.zst archives. Source connectors exist for BookStack and Confluence.

Key features

Typed knowledge graph with chunk-level evidence and cited answers
Native MCP server with 23 tools, scoped per IAM token
Eight LLM providers in the catalog, with per-call USD cost tracking
Docker Compose stack and a Helm chart for Kubernetes
Code-aware ingest using tree-sitter parsing for 15 languages
Restart-safe processing and tar.zst backup and restore

When to use it

Giving an agent grounded, cited answers from an internal document library
Indexing a codebase together with its docs and configuration files
Syncing a BookStack or Confluence space into a searchable knowledge base

Who it is for: Teams and developers who want a self-hosted knowledge base that agents can query over MCP.

How it fits with Hermes Agent

IronRAG is a general knowledge system. Hermes Agent connects to it as one of several MCP clients, and the repository is tagged hermes-agent.

How to install IronRAG

These commands are copied from the project's README. Check the repository for the latest steps before you run them.

curl -fsSL https://raw.githubusercontent.com/mlimarenko/IronRAG/master/install.sh | bash

Requirements: Docker with Compose for the stack, and access to an LLM provider such as OpenAI, DeepSeek, Qwen, OpenRouter, MiniMax or Ollama.

FAQ

What is IronRAG?

IronRAG is a self-hosted knowledge memory system. It turns documents into a typed graph with chunk-level evidence and serves it to people and agents, including through an MCP server.

Does IronRAG work with Hermes Agent?

Yes. The README lists Hermes among the MCP-compatible agents and clients you can connect, and the repository carries the hermes-agent topic.

How do I install IronRAG?

Run the install.sh script from the README with curl and bash. It is an interactive wizard that inspects the host, recommends a resource profile and asks for the port and optional admin bootstrap before writing anything.