rembric: memory tool for Hermes Agent

Self-hosted MCP memory, sessions and dashboard for coding agents, in one Docker image and one SQLite file

What rembric does

rembric runs as a single Node process in one Docker image with one SQLite file. The MCP memory surface is the core, and sessions, judgments, consolidation and an operator dashboard come with it. The MCP endpoint is on port 8787 under /mcp and the dashboard under /dashboard. The image bundles Node 22, better-sqlite3, sqlite-vec, onnxruntime-node and the embedding model, built for linux/amd64 and linux/arm64, and the README asks for about 1 GB of RAM, with 2 GB recommended.

First-class plugins handle session lifecycle and per-project path scoping for Claude Code, opencode and Codex CLI. Hermes Agent has a native provider plus MCP, Pi has an npm package with its own MCP client, and ChatGPT connects as a custom connector over OAuth 2.1. Cursor, Windsurf, Claude Desktop, Cline, Goose and other MCP clients use a plain MCP URL. Content in private tags is redacted on the client before transcript text is uploaded. One installer script sets up the server, generates the admin token and installs or removes plugins.

Key features

MCP memory tools shared across agents
Sessions, judgments, consolidation and an operator dashboard
Single Docker image with one SQLite file and an in-process embedder
Native Hermes Agent provider plus MCP access
Client-side redaction of private-tagged content
Interactive installer for the server and agent plugins

When to use it

Sharing one memory between Hermes Agent, Claude Code and Codex CLI
Reviewing what agents remembered through the dashboard
Running agent memory on a home server or a Tailscale-reachable box

Who it is for: Developers who use several coding agents and want one self-hosted memory layer for all of them.

How it fits with Hermes Agent

It supports Hermes Agent among other agents, offering a native Hermes provider alongside the MCP endpoint.

How to install rembric

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/susomejias/rembric/main/install.sh | sh

Requirements: Docker and about 1 GB of RAM (2 GB recommended)

FAQ

What is rembric?

rembric is a self-hosted memory, session and dashboard service for AI coding agents. It runs as one Docker image with one SQLite file and exposes its memory over MCP.

Does rembric work with Hermes Agent?

Yes. The README lists Hermes Agent as a supported agent with a native provider plus MCP. It also supports Claude Code, Codex CLI, opencode, Pi and other MCP clients.

How do I install rembric?

Run the install.sh script from the README with curl piped to sh, then pick Server to install and Plugins to add your agent. The README also shows how to download and read the script first.