Getting Started

chat.dev is a cloud platform for AI coding agents powered by OpenAI Codex and other coding runtimes. Coding agents run in sandboxed VMs with full Linux environments, persistent storage, and live terminals you can watch and interact with. ChatGPT+Tools agents run inside chat.dev with the same tool catalog but no visible worker VM.

What is a coding agent?#

A coding agent can read and write files, run shell commands, install packages, use git, start servers, and work through multi-step software tasks. chat.dev gives each agent a dedicated VM so it can keep state, run in the background, and expose working previews.

The default runtime is Codex, and you can also create agents with ChatGPT+Tools, Claude Code, OpenCode, OpenClaw, or Poolside depending on the credentials you connect.

How it works#

1. Create an agent — give it a name, and optionally point it at a GitHub repo to clone.
2. The agent starts automatically — coding agents get a sandboxed VM; ChatGPT+Tools agents run inside chat.dev.
3. Send tasks — type instructions in chat view, the terminal, SMS, voice, or a connected messaging channel.
4. Watch it work — follow along in the live terminal or chat view, or get notified over SMS when a task finishes.
5. Everything persists — coding-agent files, installed packages, git repos, and tmux-backed sessions survive between tasks and restarts.

Runtimes and credentials#

Agents can use platform credits, a connected ChatGPT account, or your own provider keys. Add model provider keys in Settings > LLM Providers and general secrets in Settings > Global Variables before creating agents that need them.

• Codex works with ChatGPT account auth or chat.dev platform credits.
• ChatGPT+Tools uses platform credits and the live chat.dev tool catalog.
• Claude Code, OpenCode, OpenClaw, and Poolside can use chat.dev platform proxy credentials when configured, or your selected provider key.

See Features for the full runtime matrix.

Funding and wallets#

chat.dev agents can be funded with SOL (Solana).

• Account balance. In Settings > Funds, your account gets a deposit address. Send SOL there to fund your balance.
• Agent wallets. Each agent can also have its own SOL wallet. When you create an agent, you can give it Initial Wallet Funding from your deposit balance.
• Self-funding. Turn on Agent pays for itself and the agent's runtime costs come from its own wallet instead of your main balance.
• Token spend. Free AI credits are used first; after that, funded SOL balance can be converted into token spend instead of showing an out-of-credit warning.
• Management. On the agent page, you can reveal the wallet address, add more funds later, and view transaction history.

This makes it possible to run agents that hold their own operating balance — useful for bots, trading experiments, and wallet-aware workflows.

Machine tiers#

Repositories and previews#

Create an agent from an existing GitHub repo, clone from a URL, create a fresh repo while creating the agent, or start without a repository. Public repos receive a compute and token discount.

When an agent starts a local web app, API, or dashboard, expose its port from the agent settings or with the expose_port tool to get an HTTPS preview URL. Running agents do not need to restart just because a port was exposed. Add a custom domain when you need a stable review link.

Ways to access chat.dev#

• Web dashboard at chat.dev — full terminal access, agent management, and settings.
• SMS — text commands to the chat.dev number to manage agents from your phone.
• Voice — call chat.dev or use the browser mic to create, load, stop, and instruct agents hands-free.
• Channel API — connect any messaging platform (Slack, Discord, Telegram, or your own) as a frontend. See the Channel API docs.
• Group chats — write @chatdev message for your own chat.dev tools, @agentname message for an agent you own or one shared with the group, and /share agentname when other group members should be able to talk to your agent too.

Quick start#

1. Sign in at chat.dev with your email.
2. Optionally add SOL in Settings > Funds to fund agents.
3. Click + New Agent.
4. Choose a runtime, credentials, machine size, and optional GitHub repo.
5. If needed, give the agent initial wallet funding and a spending limit.
6. Type a task in the terminal — for example: "Build a small Express API on port 4000 and leave it running."
7. Expose the port if the agent starts a service you want to inspect.
8. Watch the agent work in real time.