Quickstart

This takes you from nothing to a working memcode session in a few minutes.

1. Install

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

See Installation for the Go and from-source options.

2. Connect a backend

memcode needs a backend to talk to a model. The first time you run it in a terminal, a short wizard offers the options it detects. The lowest-friction paths:

If ANTHROPIC_API_KEY or OPENAI_API_KEY is already exported in your shell, memcode uses it automatically. Nothing to set up.

If you have a Claude, ChatGPT, or Copilot subscription, reuse it:

MEMCODE_CREDENTIAL_SOURCE=claude memcode

If you want a hosted account with one balance across vendors:

memcode login

Full details are in Authentication.

3. Run your first task

Change into a git repository and give memcode a task:

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

memcode orients itself, proposes and runs the tool calls it needs, and in the default ask mode pauses for your approval before anything risky.

4. Or work interactively

Run memcode with no arguments for an interactive session:

memcode

Type tasks in plain language. Slash commands steer as you go: /plan to plan a change before making it, /model to switch models, /mode to change how much it does on its own, and /cost to see what the session has spent. The full list is in Slash commands.

5. Come back later

Your work is remembered. Continue the most recent session with its full history:

memcode --continue

See Memory for resuming a specific session and everything else memcode keeps.