What is Memcode? More Than a Coding Agent, in Your Terminal

Memcode is an AI coding agent that runs in your terminal: you tell it what you want in plain language, and it reads your code, runs commands and edits files, asking before anything risky. It is also more than a coding agent. The same program can answer you on Slack, Discord, email and other channels, run work on a schedule, review pull requests and build websites.

Memcode is open source under the MIT license, runs on macOS, Linux and Windows, and works with the model provider you prefer, including subscriptions you already pay for.

The problem Memcode solves

AI coding tools are useful, but they come with three common frustrations.

You re-explain your project every session. Most agents start from zero each time. They re-read files and re-ask questions you answered yesterday, and you pay for that warm-up.

You are tied to one vendor. When a better model comes out from a different company, switching often means a different tool and a different workflow.

The agent only works when you are at your desk. Many useful jobs, like a morning summary, a dependency check, or a fix you think of on the train, happen away from the terminal.

Memcode addresses each one. It remembers your work, it runs on many providers, and it can live where you already talk.

What Memcode can do

Memcode is one core with several products on top. Write the code, review it, run agents on it, and ship it.

A coding agent that knows your repo

Run memcode in a project and describe the task. For anything bigger than a quick edit, /plan researches your repository without touching it, shows a step-by-step plan, and only makes changes after you approve.

You decide how much it does on its own. The default mode asks before writing files, installing, or running risky commands. Auto mode handles routine work without interrupting you. A short list of destructive actions, like wiping files, force-pushing Git history or dropping a database, always asks first, and you can't turn that off.

If you change your mind, /rewind undoes the agent's edits back to an earlier point.

Memory of your project

Memcode remembers your project and your past sessions. You can stop, come back later with memcode --continue, and pick up where you left off. You can ask about past work with memcode recap or memcode recall, give it standing rules in a MEMCODE.md file, and give it background facts in a memory.md file.

Memory is one feature, not the whole product. It saves you from re-briefing the agent every morning.

Agents in the channels you already use

Memcode can run as a gateway: an always-on assistant that answers on the channels you use and turns each message into real work in your projects. It supports twelve channels:

• Slack, Discord, Telegram and WhatsApp
• Email and SMS
• Signal, Matrix and Mattermost
• Microsoft Teams and Google Chat
• GitHub

Send a voice note on Telegram, WhatsApp, Signal or by email and it is transcribed and treated as the task. You decide who can talk to it: the gateway answers no one until you allow them, and new people pair with a short code you approve.

Each agent can have its own identity, memory and skills, so your researcher is not your support bot. You can even set agents up by conversation with memcode admin.

Work that runs on a schedule

Schedules run a task on a timer and deliver the result to a conversation, like a weekday standup summary in Slack. You can ask in plain language too: "every Monday morning, check this repo's dependencies for updates and tell me what changed". Check on those tasks with memcode task list and memcode task inbox.

An agent can also be given a standing objective and permission to work on it unattended. It works only under a policy you approved, keeps a record of what it did, and pauses to ask you when it needs something only you can answer.

Code review, websites and chat

• Code Review connects to a repository through a GitHub App and reads every pull request against the whole codebase. It stays quiet on a clean pull request and comments when something is actually wrong.
• Websites turns a description into a real static site at your own memcode.app address, with custom domains supported. It is in preview.
• Chat puts every model in one browser thread, grounded in your own documents.

Plays well with your other tools

• MCP, both ways. Memcode can use Model Context Protocol servers as tools, and memcode mcp serve turns Memcode into an MCP server that lets other tools ask what it knows about your codebase.
• Skills. Reusable playbooks for specific tasks. Memcode picks them up automatically, can install them from the skills.sh ecosystem, and also finds skills you already have in .claude/skills.
• Hooks, code navigation, and image and PDF input are included.

Models and providers

Memcode works with OpenAI, Anthropic, Google, xAI, Kimi, GLM and more, plus local models through Ollama. Current models include GPT-6 Astra, the GPT-5.6 family, Claude Fable 5.1, Opus 5.5, Sonnet 5, Haiku 4.5, Gemini 3.1 Pro, Gemini 3.8 Flash, Grok 4.7, and GLM, Qwen, Kimi and DeepSeek models through Fireworks.

You pick one model with /model and the whole session runs on it. Memcode never swaps it out mid-task, and it remembers your choice per project.

There are four ways to connect:

1. A subscription you already pay for. Claude Pro or Max, ChatGPT, GitHub Copilot, or Grok through SuperGrok or X Premium+. Run memcode auth use claude, codex, copilot or grok, at no extra cost.
2. Your own API key. If ANTHROPIC_API_KEY or OPENAI_API_KEY is set, Memcode uses it automatically.
3. A hosted Memcode account. One balance across providers, a place to store your own keys, and hosted web search.
4. A custom endpoint. Any OpenAI-compatible server, such as Ollama or vLLM.

How Memcode compares to Claude Code and OpenAI Codex

Claude Code and Codex are well-known coding agents. Memcode is designed to work alongside them, not force a choice.

Here is what Memcode brings in one place:

Products change often, so check each tool's own docs for its current features.

What Memcode is not

• Not a chat wrapper. It reads, runs and edits in your real project, with your approval.
• Not an IDE plugin. It is a single program in the terminal you already have open. No Electron, no browser tab to babysit.
• Not locked to one vendor. Switch models without rebuilding your workflow.
• Not an agent that acts without limits. Risky actions ask, destructive ones always ask, and unattended agents work only inside permissions you granted.

Pricing

The Free plan runs the full product, the CLI and the web app, on your own API keys, with no card. Paid plans add hosted models, so no keys are needed:

Model usage on paid plans comes from prepaid credits, billed at the provider's exact cost with no markup. New paid plans start with a one-time $5 credit. Turns served on your own keys are never debited. Details are on the pricing page.

Frequently asked questions

Is Memcode only for coding?

No. Coding is one use. Memcode also runs agents in your chat channels, works on schedules, reviews pull requests, and builds websites.

Can I use my ChatGPT or Claude subscription?

Yes. Memcode can run on Claude Pro or Max, ChatGPT, GitHub Copilot, or SuperGrok and X Premium+. Pick one with memcode auth use.

Does Memcode work with local models?

Yes. Point it at Ollama with memcode ollama followed by a model name, or at any OpenAI-compatible server. A local setup needs no Memcode account.

Is Memcode open source?

Yes. The CLI, agent and memory are open source under MIT, and the code is on GitHub.

What platforms does it run on?

macOS, Linux and Windows, on amd64 and arm64.

How to get started

Install Memcode with one line:

curl -fsSL https://memcode.ai/install.sh | sh

Or install with Go:

go install github.com/memcode-ai/memcode@latest

The first time you run it, a short wizard finds subscriptions you already have and offers them. Then change into a project and give it a task:

cd ~/code/my-project
memcode run "add a health check endpoint at /healthz that returns 200"

Or run memcode on its own for an interactive session.

Where to go next:

• Quickstart
• Installation
• Connect a model
• Agents and channels
• Websites
• Supported models
• Pricing