Memtro CLI
memtro: a coding agent in your terminal that reads, edits and runs commands in your repository, with your company's memory and connected systems as its tools.
Type memtro in a repository and a session opens there, full screen in a terminal, with the conversation rendered as Markdown above the input and a menu that narrows as you type a / command or an @ file name. The agent loop runs on your computer with Memtro as the model: it reads, lists, searches and edits files and runs commands in that directory, and it can search your organisation's memory and, with actions turned on, act in your connected systems. Apart from edits inside the repository and a short list of read-only commands, every command, write and action is judged first by a safety check, a separate call to a model, or put to you, and some things (sudo, piping a download into a shell, force-pushing, anything touching credentials) are always asked. There is no sandbox: commands run as you. The first run signs you in through the browser, with a key that acts as you in one organisation for 90 days. memtro run "<task>" does one task for a script and exits, asking nobody, and the one-shot commands (ask, search, remember, why, tools, call) are still there. It installs as one standalone executable, with no Node.js (that is needed only to build from source), on macOS, Linux with glibc and Windows through WSL, and memtro update checks each release's build attestation before it swaps the new file in.
What you get
- Local tools: read, list, glob, grep, edit, write and bash, run in the directory you started in
- Memtro's memory and connected systems as tools; actions in those systems only with /actions on or --actions, and only once an administrator has granted them
- Two modes: auto (a safety check judges commands and edits, and asks you when unsure) and ask (every edit, command and action is put to you); an always-asked list neither mode can loosen
- A full-screen session with a status line, a predictive / command menu, an @ file menu, Markdown answers with tables and highlighted code, dark and light themes, and /copy; --inline for the scrolling interface
- Sessions saved per directory, with --continue, --resume, compaction, slash commands, and AGENTS.md or CLAUDE.md read as project instructions
- memtro run for scripts and CI, with --json events and exit codes; MEMTRO_API_KEY in CI
- Browser sign-in to a 90-day key listed on Profile → API keys; revoke it there or with memtro logout
- Updates with /update or memtro update, which check the release's SHA-256 and build attestation; plain http:// Memtro addresses only on the local machine
How to connect
- 01curl -fsSL https://raw.githubusercontent.com/memtro-org/memtro-cli/main/install.sh | sh
- 02cd into a repository and run memtro; approve this computer in the browser the first time
- 03Say what you want done, and answer y, a, n or e when you are asked
Example queries
- › memtro run "make the failing test in src/cart pass"
- › git diff | memtro run "review this"
- › memtro ask "what did we agree with Baltic about the renewal?"
- › memtro search "deploy freeze" --json
Why it is better with memory
Through Memtro CLI you get the same memory, connector tools, routing and jobs as every other client, so what one person learns in one tool is available to the whole team in all of them.