Memryzed

Persistent memory and session state for AI coding agents.

Memryzed is a local-first MCP server that gives any MCP-aware coding agent (Claude Code, Kiro, Codex, Cursor, Copilot CLI, Continue, and others) durable memory across sessions and resumable working state per repository.

Your agent stops forgetting. Your work stops restarting from zero.

See it work

Tell your agent something once:

Ask later, in a new conversation — it remembers:

Related MCP server: pathmark

Status

Pre-1.0. Memryzed is feature-complete against the v1 specification (docs/specs/v1.md): local memory store, hybrid retrieval, per-repository sessions, the rule-based extractor with an optional local-LLM extractor, and an MCP server exposing nine tools. The API and on-disk format may still change before 1.0; see docs/development/versioning.md for the compatibility policy and CHANGELOG.md for what has shipped.

Install

macOS, Linux, WSL:

curl -fsSL https://memryzed.com/install.sh | bash

Windows PowerShell:

irm https://memryzed.com/install.ps1 | iex

Windows Command Prompt:

curl -fsSL https://memryzed.com/install.cmd -o install.cmd && install.cmd

After install:

memryzed init        # one-time setup
memryzed install     # auto-wire into detected MCP clients
memryzed doctor      # verify everything is working

For detailed setup, see docs/getting-started.md.

What Memryzed gives you

Three layers of persistent context, exposed to any MCP-aware agent through a small, stable tool surface:

• Global memory. User-wide preferences and facts that follow you across every project and every machine.

• Project memory. Repository-scoped facts that persist across sessions in the same repository: build commands, conventions, ownership, decisions.

• Session state. Per-task working state that lets you resume exactly where you left off, including open files, recent conversation, and last commands.

All data lives on your machine in a single SQLite database. No accounts, no telemetry by default, no network calls.

How it works

Memryzed runs as a local MCP server that your agent talks to over stdio. Agents call nine tools to recall facts and conversations, store facts, checkpoint sessions, and resume them. A background engine captures your agent conversations and proposes facts; you approve, edit, or reject the proposed facts through a CLI review queue.

For the full architecture, see docs/architecture.md.

Documentation

User documentation:

• docs/getting-started.md - install and first use

• docs/concepts.md - the memory model and scopes

• docs/cli-reference.md - every command

• docs/mcp-reference.md - every MCP tool

• docs/configuration.md - configuration options

• docs/troubleshooting.md - when something is wrong

• docs/faq.md - common questions

For agent and client authors:

• docs/for-agent-authors.md - how to integrate Memryzed cleanly

For contributors and operators:

• docs/architecture.md - system architecture

• docs/data-model.md - on-disk format and schema

• docs/specs/v1.md - the full v1 specification

• docs/development/ - development, release, and incident-response process

• docs/roadmap.md - what is planned next

License

Apache-2.0. See LICENSE and NOTICE.

Contributing

See CONTRIBUTING.md for the development setup, branching model, and how to propose changes. By participating in this project you agree to the CODE_OF_CONDUCT.md.

Security

To report a vulnerability, follow the process in SECURITY.md. Do not open public issues for security reports.

Authors

Memryzed was created by Hamza Arjah, founder and lead maintainer. See AUTHORS for the full attribution and the repository's contributors page for everyone who has contributed.