Use Claude Code, Codex and other coding agents in one app
Tesseract is a free desktop app for Mac, Windows and Linux that runs 14 coding agents, Claude Code, Codex and Gemini CLI among them, in one window. You install the agents you want in Settings → Models & accounts → Frameworks, sign in with the plan or key you already have, and pick one for each conversation. Two conversations can run on two different agents at the same time, and you can move a conversation to another agent without losing what was said.
Checked against Tesseract 0.12.5.
How do I use a coding agent like Codex or Gemini CLI in a desktop app?
Install Tesseract, let it install the agent, sign in, and type what you want in a conversation. The agent does the work; Tesseract puts a code editor, a terminal, a live preview and a plan you approve around it.
- Download Tesseract. It is free, and it installs and updates itself.
- Open Settings → Models & accounts. Under Frameworks, press Install next to the agent you want. The list opens on the agents already on your computer; Configure additional frameworks shows the rest. Details shows the version, the install command and the install’s output.
- If an install fails, Tesseract offers to fix it: press Fix it for me, and its assistant works out why and fixes it while you watch every step, with Stop to end it. It is free, with a daily limit.
- Sign in. In Settings → Models & accounts, press Add an account, choose what it runs on and which framework, then fill in what that pair needs. It can run on your subscription, your API key, or the agent’s own sign-in already on your computer.
- Open your project folder and start a conversation. New conversations start on the framework set in "New conversations start on", under Frameworks.
- To use a different agent in this conversation, click the model button beside the message box, open Runs on and pick the agent. This is there for everyone; you do not need Expert mode. Without Expert mode, the button shows what the model is for, such as Recommended or Fast; with Expert mode on, it shows the model’s own name and its thinking level.
Already use Codex, OpenCode, Gemini CLI, Cursor, Grok, Kimi, Qwen Code, CodeBuddy, MiniMax or Hermes on their own? The conversations you started there show up in Tesseract’s conversation list and search, next to your Claude Code ones, and you can carry one on with everything said so far.
Which coding agents can I use?
Fourteen, all from one menu. Each agent’s own page says how to sign in and what works with it in Tesseract, feature by feature:
- Claude Code, by Anthropic
- OpenAI Codex, by OpenAI
- OpenCode, open source
- Gemini CLI, by Google
- Antigravity CLI, by Google
- Kimi Code, by Moonshot AI
- Qwen Code, by Alibaba’s Qwen team
- MiniMax Code, by MiniMax
- CodeBuddy, by Tencent
- Grok Build, by xAI
- Cursor, the agent from Cursor as a command-line tool
- Hermes Agent, by Nous Research
- OpenClaw, open source
- NanoClaw, open source, built on the Claude Agent SDK
Which one should I start with?
Start with the one your account already pays for.
- You pay for Claude Pro or Max: pick Claude Code and sign in with your Claude account.
- You pay for ChatGPT: pick Codex and sign in with ChatGPT.
- You have a Google account: pick Antigravity CLI, which takes personal Google accounts and Google AI Pro and Ultra plans. Gemini CLI takes a Gemini API key, and its Google sign-in now works only with a Gemini Code Assist licence: since June 18, 2026, personal Google accounts and Google AI Pro and Ultra plans no longer work in it.
- You have no AI account: a new computer starts on OpenCode, set up for you, which works with the models included in Tesseract, on the free allowance.
You can change your mind at any point: the choice is per conversation.
Do I sign in with my own plan or an API key?
Either, depending on the agent, or neither if you use Tesseract’s free allowance. Tesseract itself charges nothing; what the agent uses counts against your plan’s own limits, or is billed to your key.
- Your plan or account: Claude Code (Claude Pro or Max), Codex (ChatGPT) and Antigravity CLI (Google). Kimi Code, Grok Build and Cursor use your account with their maker, Qwen Code and MiniMax Code your Qwen or MiniMax sign-in, and CodeBuddy its own sign-in on Tencent’s international site, each set up in the agent itself.
- An API key: Claude Code (Anthropic), Codex (OpenAI), Gemini CLI and Antigravity CLI (Gemini), Hermes Agent and OpenClaw (OpenRouter), and NanoClaw. Qwen Code takes an OpenAI-compatible key and MiniMax Code a MiniMax key, set in the agent itself. OpenCode logs in to any of its 75+ model providers, one by one.
- No account of your own: frameworks can also run on the Tesseract Gateway, where Tesseract hosts the models and there is nothing to set up.
Add more than one account and they back each other up: when one reaches its limit, the next takes over. When your Claude plan’s limit stops a reply, the chat says "Your Claude plan’s limit is reached", says when it resets, and offers Use another framework. When the framework or model answering changes, for example because another of your accounts takes over, a short note in the conversation says what answers now and why.
What does Tesseract install for me, and what does it reuse?
Tesseract installs the agents you pick, and if one is already on your computer it uses that copy. The two exceptions are Antigravity CLI, where Tesseract installs Google’s own server for apps and does not use an agy command already on your computer, and NanoClaw, which Tesseract runs on an adapter for the Claude Agent SDK.
| How it installs | Agents |
|---|---|
| With npm, so Node.js must be on your computer | Codex, Gemini CLI, Kimi Code, Qwen Code, MiniMax Code, CodeBuddy, Grok Build, Hermes Agent, NanoClaw |
| With the maker’s own installer | Claude Code, Cursor, and OpenClaw on Windows |
| Without needing Node.js | OpenCode (a ready-built program), and OpenClaw on a Mac or Linux, which brings its own Node.js if your computer has none |
| From Google’s own release | Antigravity CLI |
When Gemini CLI runs on the Tesseract Gateway, it runs in a separate, clean setup, and your own Gemini CLI and its Google sign-in are untouched.
How do I pick the agent a conversation uses, and switch it?
Click the model button beside the message box, then Runs on: it changes this conversation only. It lists each framework once, with its logo: the ones installed on the computer the conversation runs on, plus OpenCode. When you have two or more accounts for that framework, Run on picks the account, from the next message. New conversations still start on the defaults in Settings.
What happens when you switch:
- If the agent is working, that turn stops at once, and the conversation moves with everything up to the stop.
- A line above the message box says where the conversation is going and how many messages it carries; anything you type meanwhile waits and goes there once it arrives.
- The new agent is told which commands ran, which files were touched and what the latest steps returned. For a very long conversation, the most recent part goes along in full, and everything before it is kept word for word in a file the new agent can read.
- The new agent has Tesseract’s tools from its first answer, so it can still ask you a question, show a form or keep the plan.
- The model list then offers only models the new framework can run.
A framework that can’t pick up this conversation says "starts fresh" beside its name. You can also just ask: a conversation can move itself, or another conversation, to another framework or model. Opening a past conversation that ran on another framework switches to that framework, ready to continue. A fork keeps its conversation’s framework, account and settings, and you can merge conversations that ran on different frameworks.
Can two agents work on the same project at once?
Yes. Each conversation has its own framework and model, so Claude Code can work in one while Codex or OpenCode works in another, in the same project or in different ones.
How their changes are kept apart:
- They know about each other. Conversations in the same project can see each other, tell whose each uncommitted change is, and send each other a word. "Working here", at the top of Conversations, shows who else is in the project.
- You can see whose change is whose. "Uncommitted changes", under the file tree, shows which conversation made each change.
- They do not edit each other’s files. Each session claims the files it is changing, so another session cannot edit them underneath it.
- Each commits only its own work. Each session commits or undoes only its own changes, so one no longer commits another’s half-done work.
- Agents outside the app count too. Claude Code running in a terminal or an editor beside them is part of this, and Codex, Gemini CLI, Cursor or an editor’s agent can join through the tesseract-crew tool server.
A conversation can also run a team: it starts other full conversations for parts of the work, such as a developer, a tester and a reviewer, on any framework, follows them and brings their work together.
How do helpers (subagents) work with different agents?
A helper is a separate agent the conversation hands one part of the work to. On Codex, OpenCode, Gemini CLI and the other frameworks besides Claude Code, the agent can hand work to a helper on another installed framework or model. Claude Code has subagents too, listed under "What works" on its page.
A helper can be:
- read-only, so it looks but changes nothing;
- on its own branch: it works in its own Git worktree (a separate working copy of your project), and its changes wait on a branch for you to review;
- in the background, and continued later;
- one of many run at once as a workflow, with a budget and resume.
To use one, ask for it in your message, for example "have a helper on Codex review this, read-only". The activity map, from the button beside the message box, shows each helper and what it is doing, and lets you pause, resume or stop it. What a helper spends counts in its conversation’s cost, and toward a spending limit set per conversation.
Questions
Is it free to use several agents in Tesseract?
Yes. Tesseract has no subscription; each agent uses your own plan or key, or the free allowance through Tesseract.
Can I use my Claude and my ChatGPT subscription in the same app?
Yes. Sign in to Claude Code with your Claude account and to Codex with ChatGPT, then pick either one in each conversation.
Do I lose the conversation when I switch agents?
No. Runs on carries everything said so far to the new framework, including which commands ran and which files were touched.
Where do I install a new agent?
In Settings → Models & accounts, under Frameworks, where each agent has its own Install, Update and Uninstall button.
Can two agents edit the same project without breaking each other’s work?
Yes. Each claims the files it is changing, and each commits or undoes only its own changes.
I use the Codex app already. Will Tesseract see those conversations?
Yes. Conversations from the Codex app appear in Tesseract’s list and search, and you can carry one on.
More guides
- How to add an MCP connector (MCP server) to TesseractAdd any MCP server to Tesseract by its address, its command or its JSON settings, or from the official registry. Every agent framework can use it.
- Does the AI agent run on my computer or in the cloud?The agent runs on your computer, or on your other computers and servers. The AI model runs at the provider you pick, unless you use a local model.
- What does an AI task cost in Tesseract?Tesseract is free. Your Claude or ChatGPT plan, an API key or the free allowance runs the AI. Pictures and video use paid models.