Your AI coding agents,
on a server you own.
chatcode.dev is a browser terminal that opens persistent AI coding sessions on your own VPS. Bring the Claude, ChatGPT or Gemini subscription you already pay for. Pick up where you left off – from a laptop, a phone, or a Telegram chat – without juggling SSH keys or losing context.
A small piece of glue between
your browser and your server.
You bring a Linux box. We give you a clean way to run AI agents inside it – visible from anywhere, with the kind of session persistence that ssh + tmux only sort of gives you.
Browser terminal
A full xterm-compatible terminal in a tab. Drag-and-drop file upload, scrollback, copy/paste – all the things you'd expect.
Persistent sessions
Each session is tmux-backed. Close your laptop, walk to a café, reopen – your agent is still there, still mid-thought.
Your VPS, your data
Code, secrets and tool state all live on the server you control. We don't store your repos and we don't keep your SSH private keys.
Bring your own AI
Works with the Claude Code, Codex, Gemini and OpenCode CLIs – signed in with your own account. We don't resell tokens.
Continuity across devices
Same session, same agent, same context – whether you opened it on the desktop, a phone, or as a Telegram conversation.
Sandbox toggles
Per-session sandbox switch for Claude, Codex and Gemini. You decide whether the agent gets full access or stays in a box.
From signup to a live session in a few minutes.
Four steps and one of them is “wait roughly thirty seconds for your droplet to boot.”
Sign in with your email
Magic-link auth. No password to remember, no extra account to manage.
Hook up a server
Easy mode: connect DigitalOcean and we'll spin up a droplet for you. Manual mode: paste a one-line installer onto any Linux box and the gateway connects itself.
Open a session
Pick a workspace folder under ~/workspace,
pick an agent (Claude, Codex, Gemini, OpenCode) and you
land in a real shell. Sign into the agent once, the way
you would on your laptop.
Keep it running
Sessions stay alive between visits. Drop files in by drag-and-drop, get push notifications when long jobs finish, and reattach later from anywhere.
The CLIs you already use, preinstalled.
On a fresh server, the gateway sets up Claude Code, Codex CLI, Gemini CLI and OpenCode under your user, ready to log in.
One session, every device you carry.
The browser is the main view, but you don't always have the browser. Connect a per-user Telegram bot and the same session shows up as a chat – paste a screenshot, ask the agent a follow-up, get pinged when it's done.
-
Same
session_idacross web and Telegram. - Notifications from the agent when long jobs land or stall.
- Mini-app pages for picking a server and resuming a session.
- Auto-suspend when the page is hidden, instant restore when you come back.
What runs where, in plain words.
We'd rather be specific about the trade-offs than wave hands. The short version is below; longer security notes live in the repository.
Your code, your VPS
Files, environment variables and tool state live on the box you provisioned. We don't store your repos and we don't keep your SSH private keys.
TLS on every hop
Browser ↔ control plane and control plane ↔ gateway both run over TLS. The control plane relays terminal traffic – so today, it can see it. We're upfront about that and a payload-encrypted mode is on the roadmap.
Pinnable releases
Gateway releases ship with checksums and explicit version
tags. Pin --version vX.Y.Z instead of
latest when you want predictable installs.
Honest MVP scope
Today there's no end-to-end encryption between browser and gateway, and free plans are capped at 10 concurrent sessions per VPS. Both are written down, not buried.
Workspace-scoped sessions
Each session is rooted under ~/workspace in a
folder you choose. Sidebar groups by workspace, tabs show
the relative subpath, and nothing leaks across.
Time-limited support access
If you grant us SSH for a debug session, it's an opt-in key with auto-expiry. You can revoke it in the dashboard at any time.
Reasonable questions, short answers.
Is it really free?
chatcode.dev itself is free to start using – there's no pricing page yet, and no credit card to sign up. The cost you do pay is for your own VPS (DigitalOcean droplet or any Linux box you already have) and for whichever AI subscription you bring along.
Do I have to use DigitalOcean?
No. DigitalOcean is the easy path because we can spin a droplet up for you with one OAuth step. If you'd rather bring your own server (Hetzner, a Mac mini, a Pi), paste the install one-liner from the docs and the gateway connects itself.
Whose AI tokens are being used?
Yours. Each agent (Claude Code, Codex CLI, Gemini CLI, OpenCode) signs in inside the session with your account. We don't proxy through our keys and we don't bill you for inference.
What happens if I close the browser?
Nothing dramatic. The session is tmux-backed, so the agent keeps running on your VPS. When the page becomes hidden the terminal connection suspends; when you come back, it restores automatically.
Can the chatcode team see my code?
Your files live on your VPS – we don't pull them. The control plane sits in a trusted relay position for terminal traffic today, which means an operator with control plane access could in theory inspect what passes through. We treat that as a real trade-off rather than a marketing footnote. End-to-end terminal encryption is on the roadmap.
How many sessions can I run?
On the free plan, the control plane caps it at
10 concurrent sessions per VPS, and the
gateway itself has a safety cap at 50. That's plenty for
most workflows; if you bump into it, get in touch.
What does “MVP” actually mean here?
A small product that does one thing reliably. The first release focuses on “create a server, open a terminal, do work without losing it.” A few things you might expect from a v1.0 – end-to-end terminal encryption, paid plans, multi-user workspaces – are deliberately not in scope yet. We'd rather ship the boring parts well than the impressive parts unevenly.
Want to try it?
Sign in with email, point it at a server, open a session. You can be talking to a real agent on your own VPS in about the time it takes to read this page.