Overview
What is Memryzed?
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. All data lives on your machine in a single SQLite database — no accounts, no telemetry by default, no network calls.
How to use Memryzed?
Install with one command for macOS/Linux/WSL (curl -fsSL https://memryzed.com/install.sh | bash) or Windows (PowerShell: irm https://memryzed.com/install.ps1 | iex). After install run memryzed init for one-time setup, memryzed install to auto-wire into detected MCP clients, and memryzed doctor to verify everything works.
Key features of Memryzed
- Global memory — user-wide preferences across every project and machine.
- Project memory — repository-scoped facts that persist across sessions.
- Session state — per-task working state to resume exactly where you left off.
- All data stored locally in a single SQLite database.
- Nine MCP tools for recalling, storing, and managing facts and sessions.
- Background extraction engine with CLI review queue to approve or reject proposals.
Use cases of Memryzed
- An agent remembers your preferred test framework across new conversations.
- Resume an interrupted coding task with open files, recent conversation, and last commands intact.
- Store repository-specific conventions (build commands, coding style, ownership) that persist across sessions.
- Approve or reject proposed facts from agent conversations via a simple CLI queue.
FAQ from Memryzed
What data does Memryzed store and where does it live?
All data is stored in a local SQLite database on your machine. No accounts, no telemetry by default, and no network calls are made.
Which coding agents work with Memryzed?
Any MCP-aware agent works, including Claude Code, Kiro, Codex, Cursor, Copilot CLI, Continue, and others.
How does Memryzed extract facts from conversations?
A background engine captures agent conversations and proposes facts using a rule-based extractor, with an optional local-LLM extractor. You approve, edit, or reject proposals through a CLI review queue.
What is the current status of Memryzed?
Pre-1.0. It is feature-complete against the v1 specification but the API and on-disk format may change before stable release.
Does Memryzed require any external services or accounts?
No. It runs as a local MCP server over stdio and does not require accounts, telemetry, or external network calls by default.
