Published Updated
In one or two sentences, before anything else.
T3 Code is a free, MIT-licensed app from T3 Tools Inc., credited to Theo Browne's team, that runs the coding agents you already pay for, such as Claude Code, Codex, Cursor and OpenCode, side by side from a desktop, web or phone interface. First released on March 7, 2026, its GitHub repository had 28.1k stars on October 17, 2026.
What it can do, one sentence each, and whether we actually tested it.
Runs Codex, Claude Code, Cursor, Grok Build, OpenCode and Google Antigravity, plus Pi, Muse Code and ACP Registry agents such as Devin, Cline and Kimi.
Per T3 Code's README and v0.1.0 release notes. Each agent must be installed and signed in on the machine running the T3 server; Antigravity needs only a Google sign-in.
Each task is a thread that runs in your current checkout or a new Git worktree, with per-turn diffs, checkpoints and message-level revert.
Per Flavio Copes's guide (October 16, 2026). Worktrees separate files, not ports or databases.
One button commits a thread's changes, pushes the branch and opens a pull request.
Per T3 Code's homepage and Better Stack's March 2026 walkthrough.
With delegate_task, an agent can start child agents on another provider or model and wait for their results, for example Claude planning and Codex implementing.
Stable since v0.1.0 (October 15, 2026), per its release notes.
Fork any finished run and merge it back, or change provider, account or model between turns.
The release notes call a provider switch lossy: the new agent gets a selection of messages but not the earlier reasoning, tool calls or attachments.
iOS and Android apps and a browser client pair with the T3 server on your computer, over your own network, Tailscale, SSH or the optional hosted T3 Connect.
Per T3 Code's README, privacy policy and Flavio Copes. The server must stay running.
Uses each agent's own login, such as claude auth login or codex login; T3 Code does not resell tokens or add its own usage caps.
Per t3.codes. Your limits are those of your agent plans.
Supervised, Auto-accept edits, Auto and Full access, set per thread.
Full access is the default for new threads, per Flavio Copes and Botmonster.
The people it helps and the problems it solves, and where it ranks.
Built for
Problems it solves
Graded in
Public tests other people ran: what they gave it and how it did.
Case 1
Input
Run the T3 Code server on a Windows 11 workstation, connect it to Claude Code, and reach it privately from a phone and other machines through Tailscale Serve.
What happened
It worked after fixes: T3 tried to start Codex ("spawn codex ENOENT") until he turned Codex off, showed "No provider available" until he corrected the Claude path, and its built-in --tailscale-serve option failed on Windows, so he set up Tailscale Serve himself. After a reboot, requests queued until the Windows user logged in.
Case 2
Input
Add projects, run agent tasks in Git worktrees, review per-turn diffs, then commit, push and open a pull request in one action.
What happened
The reviewer found switching between projects and long threads instant, with no slowdown while several agents ran. At that point only Codex was fully integrated and Claude Code support was awaiting merge. A feature walkthrough, not a measured test.
Case 3
Input
Have Claude plan a change and hand the implementation to Codex through delegate_task, each running as its own child thread.
What happened
Each child keeps its provider's own history and reports a result, and the parent can wait or keep working. The notes call this the recommended way to combine harnesses, rather than switching providers mid-thread, which loses the earlier reasoning and tool calls. Documented, not tested by anyone we found.
Case 4
Input
Start with a read-only investigation of a repository, then one small edit with a diff review and a revert test.
What happened
Copes uses the current checkout for one stream of work, worktrees for independent parallel tasks, Supervised mode for valuable files and Full access only in disposable workspaces. A recommended workflow, not a reported test.
What it does badly, with the evidence, and what we could not settle.
Weak
Early, fast-changing software.
Evidence: The README says "very very early" and to expect bugs. On October 17, 2026 the repository had about 1.5k open issues and several nightly and preview builds a day; v0.1.0 (October 15, 2026) is the first stable release of the rebuilt orchestrator, and its threads are a one-way copy from the old version.
New threads get Full access by default.
Evidence: Flavio Copes (October 16, 2026) and Botmonster (September 15, 2026) both report Full access as the default. Copes notes that a permission mode is not a security boundary: full access means full access on the server machine.
Switching providers mid-thread loses context.
Evidence: T3 Code's v0.1.0 notes say the handoff is lossy: the new provider gets a budgeted selection of messages, not the previous provider's reasoning, tool calls, results or attachments.
Setup can take fixing, especially on Windows.
Evidence: Steve Smith (August 19, 2026) hit a Codex launch error, a "No provider available" screen and a failing built-in Tailscale option before it worked, and requests stalled after a reboot until he logged in.
Usage data is sent by default.
Evidence: The privacy policy (effective October 6, 2026) says T3 Code sends anonymous usage data, such as which provider and model a turn used and whether it succeeded, unless you set T3CODE_TELEMETRY_ENABLED=false. It says it does not train AI models on your code, prompts or agent replies.
Not for teams that need a finished, supported product now, or for anyone without an agent subscription.
Evidence: T3 Tools mostly does not accept contributions, has no docs site, labels features alpha or beta in its terms and caps its liability at the greater of 100 USD or what you paid. T3 Code brings no models of its own. Such teams should wait.
Uncertain
Whether hosted features will stay free.
Why unsure: The terms (July 15, 2026) say there is currently no T3 Tools subscription fee "unless we clearly state otherwise for a feature", and the hosted web app and T3 Connect are account-backed services.
Which agents can hand work to others.
Why unsure: The v0.1.0 notes recommend delegate_task for "Claude planning and Codex implementing", but their summary lists only Pi, Muse Code and OpenCode 2 as supporting it.
How many people use it.
Why unsure: The homepage says it is "tolerated by over 500,000 devs" without saying how that is counted; GitHub stars measure interest, not use.
Who exactly built it.
Why unsure: T3 Tools Inc. holds the copyright and T3 Code's site names no individuals; Better Stack credits Theo and the T3 team, and Flavio Copes credits @theo and @jullerino.
The numbers, each with its source and the date we checked it.
| Spec | Value | Note and source |
|---|---|---|
| Maker | T3 Tools Inc. (copyright holder in the MIT licence; also runs T3 Chat) | T3 Chat's terms of service name T3 Tools, Inc. as its operator. Source: github.comchecked |
| Credited to | Theo and the T3 team | Source: betterstack.comchecked |
| Licence and source | MIT; github.com/pingdotgg/t3code | Source: github.comchecked |
| First public release | v0.0.2, March 7, 2026 | Source: github.comchecked |
| Current stable release | v0.1.0, October 15, 2026, the first stable release of the rebuilt orchestrator | Source: github.comchecked |
| GitHub stars | 28.1k stars and 7.5k forks, read October 17, 2026 | Source: github.comchecked |
| Platforms | macOS (Apple Silicon, Intel), Windows 10 and 11, Linux (.deb, AppImage), iOS, Android, terminal (t3 or npx t3) | Source: t3.codeschecked |
| Price | Free; no T3 Tools subscription fee; you pay for your own agent plans | Source: t3.codeschecked |
| Agents supported | Codex, Claude Code, Cursor, Grok Build, OpenCode, Google Antigravity | Source: github.comchecked |
| Data and training | No AI training on your code, prompts or agent replies; anonymous usage data on by default, can be turned off | Source: t3.codeschecked |
Answer first, then why, an example, the limits and what to do next.
Yes, T3 Code comes from T3 Tools Inc., the company that also runs T3 Chat, and is credited to Theo Browne and the T3 team. The MIT licence names T3 Tools Inc.; Better Stack credits Theo and the T3 team, and Flavio Copes credits @theo and @jullerino.
Yes, T3 Code is free to download and use, with no T3 Tools subscription fee. You pay for the agents it runs, for example Claude Pro (20 USD a month) for Claude Code or a ChatGPT plan for Codex, and T3 Code does not resell tokens.
Limits
Its terms leave room to charge for specific features later.
Yes, T3 Code is open source under the MIT licence, at github.com/pingdotgg/t3code. The repository had 28.1k stars and 7.5k forks when we read it on October 17, 2026, up from 20,000 stars on August 27, 2026, per Flavio Copes.
Limits
The team mostly does not accept outside contributions yet.
T3 Code supports Codex, Claude Code, Cursor, Grok Build, OpenCode and Google Antigravity, per its README. Since v0.1.0 (October 15, 2026) it also runs Pi, Muse Code and ACP Registry agents such as Devin, Cline and Kimi. Each must be installed and signed in on the machine running the T3 server.
Yes, T3 Code has desktop builds for macOS, Windows 10 and 11 and Linux, plus iOS and Android apps. The phone apps are remotes: they pair with a T3 server on your computer, reached over your own network, Tailscale, SSH or the hosted T3 Connect.
Limits
Steve Smith's Windows setup needed several fixes (August 2026).
No public test has compared them on the same tasks, so the choice comes down to which agents you use. T3 Code is free, open source and runs agents from six or more vendors on macOS, Windows and Linux; Conductor runs Claude Code, Codex and Cursor on a Mac, free or 50 USD a month for Pro; Claude Code's desktop app runs parallel Claude sessions; ChatGPT's apps run Codex; Cursor is an editor with its own agents.
Next step
If you use only Claude Code, start with its own desktop app, which already runs parallel sessions in Git worktrees.
This page is built from what can be checked today: T3 Code's site, README, release notes, terms and privacy policy, its GitHub repository, and write-ups by Better Stack, Flavio Copes, Botmonster and Steve Smith (Ardalis), the last one a hands-on setup. Every date and number links to its source, and T3's claims are labelled as T3's.
T3 Code has no model of its own. It is a local server that starts the coding agents already installed and signed in on your computer, and shows them in one desktop, browser or phone interface. Its README says the team was inspired by the Codex desktop app, Conductor, Claude Desktop and Cursor Glass, "but none met our bar".
The MIT licence names T3 Tools Inc., the company that also runs T3 Chat (T3 Chat terms). Better Stack credits Theo and the T3 team, and Flavio Copes credits @theo and @jullerino; Theo's GitHub account, t3dotgg, is Theo Browne's.
| Date | Event | Source |
|---|---|---|
| March 7, 2026 | v0.0.2, the first public release on GitHub | release |
| March 8, 2026 | Better Stack walkthrough: Codex fully integrated, Claude Code pending | Better Stack |
| August 19, 2026 | Steve Smith publishes a Windows and phone setup | Ardalis |
| August 27, 2026 | Repository passes 20,000 stars | Flavio Copes |
| October 2, 2026 | v0.0.45, the last 0.0.x stable release | release |
| October 15, 2026 | v0.1.0: rebuilt orchestrator, agents that start agents, ACP Registry | release notes |
| October 17, 2026 | 28.1k stars, 7.5k forks, about 1.5k open issues | GitHub |
T3 Code is free. Its terms say there is no T3 Tools subscription fee "unless we clearly state otherwise for a feature", and the homepage says it "doesn't resell tokens". You pay for the agents it drives: Claude Code comes with Claude Pro from 20 USD a month (Anthropic), and Codex with ChatGPT plans from Free to Plus at 20 USD (OpenAI).
No public test measures T3 Code's speed or reliability against other tools.
| T3 Code | Conductor | Claude Code desktop | Codex | Cursor | |
|---|---|---|---|---|---|
| Agents | Codex, Claude Code, Cursor, Grok Build, OpenCode, Antigravity and more | Claude Code, Codex, Cursor | Claude | OpenAI | Cursor's own agents |
| Runs on | macOS, Windows, Linux; iOS and Android remotes | Mac for local workspaces | macOS, Windows; Linux in beta | ChatGPT apps, CLI, IDE | Its own editor |
| Price of the app | Free, MIT-licensed | Free; Pro 50 USD a month | With Claude Pro, 20 USD | With ChatGPT plans | Free; paid from 8 USD |
Sources: Conductor, Conductor pricing, Claude Code desktop docs, OpenAI and Cursor.
If you use only Claude Code, its own desktop app already runs parallel sessions in Git worktrees. T3 Code earns its place when you mix vendors, for example Claude planning and Codex implementing, which its release notes recommend doing through delegate_task, or when you work on Windows or Linux, since Conductor's local workspaces are Mac-only.
Sources
Changelog
Command Code is a terminal coding agent from Langbase, Inc., trading as Command Code and led by Ahmad Awais, that says it learns your coding "taste" from the changes you accept, reject and edit. Announced with 5 million USD in seed funding on February 25, 2026, it now focuses on low-cost open models, with plans from 1 USD a month.
ZCode is Zhipu AI's (Z.ai) free desktop coding-agent app for macOS, Windows and Linux, built around its GLM-5.3 model, with model use billed through the GLM Coding Plan from 18 USD a month. First released as Z Code in December 2025, it switched to Zhipu's own agent in June 2026; Z.ai said in August it had passed one million users.
VideoEnhancer is a web tool that upscales short MP4 clips to 1080p, 2K or 4K with the FlashVSR model and also compresses, face-swaps and generates video. After a 5-second free trial, 1080p costs 3 credits a second, $0.03 to $0.09 depending on how credits are bought. VideoEnhancer is our own product: it is made by the team behind GenGrades.
LayerGrab splits a flat image into up to 16 named, transparent layers over a filled-in background, on the web, in Figma or in Chrome, or pulls out only what you name. Each layer costs $0.10 on its $12 plan, $0.08 on Premium. No independent review exists yet. LayerGrab is our own product: it is made by the team behind GenGrades.
VideoTextRemover is a web tool that erases burned-in subtitles, captions and other on-screen text from MP4 and MOV videos and repaints what was behind them, up to 1080p. It bills per second processed, about $0.025 to $0.049, with 30 free seconds a month in some countries. VideoTextRemover is our own product: it is made by the team behind GenGrades.
MagicRemover is a browser tool that erases people, objects, text, logos and watermarks from photos, and subtitles from videos. Photos get 2 free removals a day, 2 more after sign-in, at about 1 megapixel; beyond that, credits start at $9.90 for 330. MagicRemover is our own product: it is made by the team behind GenGrades.