your agent,
everywhere.

me is a minimal coding agent in one Python file. Your models, API keys, prompts, skills and chats live in the cloud — install on any machine, log in, and everything is already there.

$curl -fsSL https://me-agent.ru/i | sh

v0.13.8 · latest version — already installed? run /update

Needs Python 3.8+. No pip, no node, no dependencies. Then just type me.

<1ktokens before you say hello

1Python file, stdlib only

0dependencies to install

∞any OpenAI-compatible model

three steps, any machine

No setup to repeat. Your account carries the whole environment.

install

One command from the box above. Python 3.8+ is all it needs.

log in

Type me, then /login. Models, keys, prompt, skills, themes and chats are already there.

continue anywhere

Terminal, browser tab, phone app or Telegram. The same session, live, within a second.

tiny on purpose

Most agents spend 10–20k tokens before you say hello. me starts under 1k: a sharp system prompt and a few compact tools, web search included. Everything else loads only when it's needed.

01one file

Stdlib Python. Read it, fork it, run it anywhere Python runs: Linux, macOS, Windows, a server, a Raspberry Pi.

02cloud sync

Config, providers, keys, system prompt, skills and full chat history follow your account. New laptop? me → log in → done.

03any model

Any OpenAI-compatible API: chat/completions or responses, reasoning effort, streaming thinking, custom names and context sizes.

04skills, not MCP

A skill is one .py file: instructions plus actions. About 15 tokens idle. MCP servers also work, loaded lazily the same way.

05goals and plans

A live todo checklist in every UI. /goal works until the goal is done and verified, or tells you exactly what blocks it.

06safe edits

A checkpoint per turn gives you /undo. The web UI shows a diff to keep or revert. Permissions: read-only, ask, or full.

07cloud sandbox

Start in the terminal, continue in the browser or Telegram. The session and project files follow you, isolated on the server.

08web app on your phone

me web locally, or the cloud /app as an installable PWA with attachments and push when a task is done.

09background and preview

"Do X and notify me": tasks keep running after you close the page. The site the agent built opens on your phone by link.

10plugins and voice

Plugins change me itself: commands, tools, keys, themes. Hold space to dictate with Groq Whisper.

11your own Telegram bot

Every account sets its own bot. The agent's notify tool sends you messages and files.

122FA if you want

Username and password by default. Attach Google Authenticator, Aegis or 1Password in one command.

see it

The same agent in the browser, on your phone and in your account. These are real screens.

Web chat: goal, live todo list, tool cards, diff review.

Sessions, modes and settings in the drawer.

Models and keys, synced to every machine.

Marketplace: skills, plugins, themes, agents.

Cloud: sandboxes, Telegram, GitHub, push, background jobs.

Themes shared by terminal, web app and site.

The terminal UI with fuzzy slash commands.

everything it can do

The short list above is only the start. Here is the full one.

agent

modes: code, design, plan, review, debug, docs, research
/goal until done and verified, exit codes for CI
live todo list, self-check before "done"
subagents with task(agent, prompt)
AGENTS.md always in context, /init
checkpoint per turn, /undo, diff review
permissions: read-only, ask, full + allow-list
context meter, /compact, /autocompact
cost per task, session and day, limits
web search and fetch built in

extend

skills: one .py file, about 15 tokens idle
plugins that change me itself
MCP servers, loaded lazily
marketplace with ratings, versions, auto-update
themes: me-dark, me-light, mono, graphite, Nord, Tokyo Night, Gruvbox…
custom agent modes as a .md file
any OpenAI-compatible model, reasoning effort

cloud

sync of config, keys, skills, themes and every chat
live sync between devices within a second
workspaces: project files with 3-way merge
isolated cloud sandboxes (bubblewrap)
background tasks, schedules, notifications
preview of the built site by link
GitHub: clone, branch, commit, push, PR
your own Telegram bot per account
web push, voice input, installable PWA
2FA, device list, self-hosting

skills are just python

The model sees one line per skill. When it needs one, it loads the guide and calls actions. Each call shows up as a live card with status, progress and logs, then collapses into a one-liner.

""" name: github description: issues & PRs on GitHub icon: color: magenta env: GITHUB_TOKEN --- Call list_prs before labeling. """ @action def list_prs(ctx, repo: str): "Open PRs of a repo" return ctx.http.get(f"…/repos/{repo}/pulls") @action(confirm=True) def label(ctx, repo: str, nums: list): "Label many PRs" for n in ctx.progress(nums, "labeling"): ctx.log(f"#{n}") …

╭─◉ github · label ─────── 2.4s ─╮ │ repo="me" nums=[12,15,19] │ │ labeling ██████████░░░░░ 2/3 │ │ › #12 │ │ › #15 │ ╰────────────────────────────────╯ ✓ ◉ github · list_prs 0.8s 3 open /skills → new from template, ask me to write one, install from URL

commands

Type / and start typing. Suggestions are fuzzy.

/goal /todo	work toward a verifiable goal, live checklist
/undo /perm	restore the last turn, set permissions
/model /provider	switch models, any OpenAI-compatible URL and key
/agent /theme	modes (code, design, plan, review…), themes
/skills /plugins	manage skills, plugins and the secrets they use
/sessions /new	resume any chat from any machine
/login /2fa /sync	cloud account, authenticator, sync now
me -p "task"	headless one-shot · `me --rpc` JSON-lines for your own UI

one agent, every screen

The same account, chats and skills in every app. Pick what fits the machine you are on.

⌨terminal

The full agent in any terminal: Windows, macOS, Linux, Termux. One install command at the top.

⑂me studio

Many agents at once, like the Codex app: each task on its own git branch, live progress, review, apply, commit, push. me-studio.exe or me studio

▣Windows app

Native window with the local agent and folder-based projects. me-desktop.exe

▤Android

Your cloud sandbox on the phone, voice and photos included. me-android.apk

◧me IDE

A VS Code–based editor with the agent built in. Portable: Windows · Linux

⧉VS Code extension

Add the agent to your own VS Code or Cursor: me-agent.vsix → Extensions → Install from VSIX.

◎web app

Nothing to install: open it in any browser, add it to the home screen as an app.

all downloads and nightly builds →

open source, and yours to run

Everything here is open: the agent, the web UI, the server and the deploy files. Anyone can run the server part on their own machine, so you never have to trust us with your data.

01read it

The client is me.py, built from plain files in src/. The server is server/ (FastAPI + SQLite). No hidden parts.

02run your own

Install the backend on any Linux box with Python, point ME_SERVER at it and your devices sync there instead of here.

03plain about data

The hosted server keeps your synced config, keys and chats unencrypted, and its operator can read them. Not okay for you? Run your own server.

04locked down

Passwords and session tokens are hashed, 2FA is optional, cloud sandboxes run isolated. Read marketplace skills before you install them.

github.com/Gl1ch5/Me-agent · report security issues through GitHub.

try it in a minute

Install, type me, and pick up the same session on your phone.

open the web app create an account source on GitHub

your agent,everywhere.