Is Glyph Hold free and self-hosted?

Yes. Glyph Hold is open source under the MIT License and can be self-hosted with Docker. It does not require hosted AI APIs, vector databases, embeddings, or paid APIs.

Available for self-hosting

Glyph Hold

A free, open-source, self-hosted memory and secrets service for AI agents. Give Hermes Agent, Codex MCP, and local assistants durable context, encrypted credentials, API keys, and audit history under your control.

View Glyph Hold Install with Docker

No hosted AI: No LLM calls, embeddings, vector DBs, or paid APIs inside Glyph Hold.
Agent ready: HTTP API, API keys, MCP connector, Hermes plugin, and SQLite FTS5 search.

Glyph Hold gives agents a shared place to remember the things that should survive the current chat: project context, procedures, preferences, decisions, and secrets that should only be revealed when explicitly requested. It is built for people looking for local AI agent memory that can be self-hosted for free.

View the project on GitHub

Why It Exists

Memory that stays local, searchable, and inspectable.

Agents are useful when they remember stable context. Glyph Hold keeps AI agent memory separate from the model, visible in a dashboard, and reachable through a simple API.

Durable memories

Store notes, decisions, procedures, project facts, people, preferences, and temporary context with categories, tags, confidence, and revision history.

Encrypted secrets

Store operational secrets encrypted at rest and reveal values only through explicit reveal endpoints. Secret values stay out of memory prefetch.

Agent access

Create scoped API keys from the dashboard, then connect agents over HTTP or through the standalone Glyph Hold MCP server and Hermes memory provider.

Simple self-hosting

Run it with Docker on port 5995. Data lives in SQLite, search uses SQLite FTS5, and the dashboard handles first-run setup.

Use Cases

Self-hosted memory for AI agents, assistants, and toolchains.

Glyph Hold is designed for developers and power users who want agent memory without handing private context to a hosted service. Run it locally, connect agents with API keys, and keep the memory database and encryption key in your own environment.

Hermes Agent memory

Use the Hermes memory provider plugin to make Glyph Hold the external memory source for Hermes Agent, with prefetch and explicit memory tools.

Codex MCP memory

Connect Codex and other MCP clients through the Glyph Hold MCP server for searchable memories, revision restore, and managed secret access.

Local assistant memory

Give custom agents and local assistants a deterministic HTTP API for durable project context, user preferences, procedures, and operational notes.

Free open-source hosting

Run Glyph Hold with Docker and SQLite. No hosted AI service, vector database, embedding provider, or paid API is required by Glyph Hold itself.

Current Status

Glyph Hold is ready for real self-hosted use.

Glyph Hold is Docker-ready and usable today for local agent memory and secrets. Keep backups of the mounted data directory and encryption key before upgrading, especially when storing encrypted secrets.

Docker image Web dashboard API keys Memory revisions Secret reveal controls MCP connector Hermes plugin SQLite FTS5 search

Install

Run Glyph Hold with Docker.

Start locally, create the first dashboard account, then generate an API key for your agent.

One-line start

GLYPHHOLD_KEY="$(openssl rand -hex 32)" && docker run -d --name glyphhold --restart unless-stopped -p 5995:5995 -v glyphhold-data:/data -e GLYPHHOLD_DB_PATH=/data/glyphhold.sqlite -e GLYPHHOLD_ENCRYPTION_KEY="$GLYPHHOLD_KEY" ghcr.io/dosk3n/glyphhold:latest && printf 'Save this Glyph Hold encryption key: %s\n' "$GLYPHHOLD_KEY"

Docker Compose

services:
  glyphhold:
    image: ghcr.io/dosk3n/glyphhold:latest
    container_name: glyphhold
    ports:
      - "5995:5995"
    volumes:
      - ./data:/data
    environment:
      GLYPHHOLD_DB_PATH: /data/glyphhold.sqlite
      GLYPHHOLD_ENCRYPTION_KEY: change-this-to-a-long-random-value
    restart: unless-stopped

Open http://localhost:5995.
Create the first dashboard username and password.
Create an agent API key from the dashboard.
Connect your agent directly over HTTP, MCP, or the Hermes plugin.

Agent Connectors

Use Glyph Hold from Codex, Hermes, and other agents.

Glyph Hold stays focused on the core service. Agent integrations live in separate repos and connect through the public HTTP API, so they can be updated independently.

Main project github.com/Dosk3n/glyphhold MCP connector github.com/Dosk3n/glyphhold-mcp Hermes memory provider github.com/Dosk3n/hermes-memory-glyphhold

Hermes

Make Glyph Hold the external memory provider for Hermes Agent.

The Hermes plugin lets Hermes prefetch relevant Glyph Hold memories and use the full memory and secret tool surface, including revision restore and explicit secret reveal actions.

Memory prefetch Create and update Archive and delete Revision restore Secret metadata Explicit reveal

Roadmap

What is coming next.

Now

Docker setup, dashboard, memories, secrets, API keys, MCP, and Hermes.

More connector hardening, stronger recovery flows, clearer docs, and broader agent examples.

Then

More integrations and operational polish for dependable self-hosted agent memory.

FAQ

Questions people ask about AI agent memory.

Can Glyph Hold provide memory for Hermes Agent?

Yes. The Hermes memory provider plugin connects Hermes Agent to Glyph Hold through the public HTTP API for memory prefetch, memory management, and explicit secret reveal tools.

Can Glyph Hold work with Codex?

Yes. Codex can use Glyph Hold through the standalone MCP server, giving Codex access to memory search, prefetch, create, update, archive, revision restore, and secret tools.

Is Glyph Hold free to self-host?

Yes. Glyph Hold is open source under the MIT License and runs with Docker using SQLite for storage and SQLite FTS5 for search.

Does Glyph Hold use hosted AI APIs?

No. Glyph Hold does not call hosted LLMs, embeddings, vector databases, Ollama, or paid APIs. It stores and searches durable memory deterministically.