Skip to main content

Decision brief

T3 Code Review: One Open-Source Control Surface for the Coding Agents You Already Pay For

Who T3 Code fits, what it adds over each agent's own app, and what a team must settle before it runs Claude, Codex and other agents side by side through it.

Reading time
5 min read
Sources
19 cited
In this piece
T3 Code+20 Positive

GILD verdict

Choose T3 Code when your engineers run agents from more than one vendor and want them in one app, and someone will set its permission default and account rules before rollout.

The workflow
Engineers running Codex, Claude Code and other agents side by side, in parallel worktrees, from a desktop, a browser or a phone, on their own plans and logins.2810
The decisive evidence
  • DocumentedT3 Code's README calls it an agent harness control surface that works with your subscriptions on Claude Code, Codex, Cursor, Grok Build, OpenCode, Google Antigravity and fx. Its FAQ says it is not a harness itself and does not work without one.24
  • DocumentedOne prompt can go to several models at once, each in its own thread and git worktree. An agent can start a sub-agent on any supported provider through T3 Code's orchestrator MCP server, and the child receives only the task prompt.89
  • DocumentedNew threads start in Full access, the most permissive of four modes. Telemetry is on by default, carries provider, model and hashed account identifiers, excludes prompts and file contents, and turns off with T3CODE_TELEMETRY_ENABLED=false.712
  • DocumentedT3 Code drives Claude through Anthropic's Agent SDK and says it is within Claude Code's terms. Anthropic's legal page says products built on the Agent SDK should use API keys and that third-party developers may not route requests through plan credentials on behalf of their users.41718
  • GILD analysisT3 Code beats each vendor's own app when a team uses several vendors' agents and wants to compare and hand work between them in one place. It loses to them when one vendor covers the team's work or when an organization needs SSO, audit logs or a support contract.
The tradeoff
You get the supported agents you are signed in to in one place at no license cost. In return you take early software that ships stable releases every few days and starts new threads in Full access, and it runs locally on each user's own agent logins.2713
What would change the call
The call changes if your Claude use must stay inside Anthropic's own apps, or if your security team requires SSO, audit logs or a vendor support contract. It also changes if the team standardizes on one vendor's agent.417

T3 Code

AI Coding Agent by T3 Tools

+20 Positive
Product page

Fit and limits

Who it fits

  • Engineers use agents from more than one vendor and want to switch, compare or hand work between them without leaving one app. T3 Code transfers conversation context when a thread switches providers.
  • You want to try one task on several models at once. Each selected model gets its own thread and git worktree.
  • Engineers want to steer agents from a phone or another machine. The iOS and Android apps act as remotes for a computer, server or VPS that runs the agents.
  • You want recurring or event-driven agent work. Scheduled tasks run on a chosen environment, and a webhook schedule can start one when GitHub opens a pull request or a CI job fails.

Where it falls short

  • It is early. The README says to expect bugs, the terms say features may be alpha or beta and may lose data, and T3 Code shipped 7 stable releases and 117 nightly builds in the 30 days to Nov 1, 2026.
  • Defaults trust the agent. New threads start in Full access, and T3 Code notes that providers enforce permission modes differently.
  • It is an MIT-licensed app users run locally with their agent logins, with support through its Discord and vulnerability reports directed to security@ping.gg, so plan any organization controls around that.
  • It does not run local models; its FAQ says so directly. Open-weight models reach it only through an agent such as OpenCode or an API-key provider instance.

Implementation reality

  1. Choose the agents and accounts

    List which agents T3 Code may drive and which login each uses: a ChatGPT plan or Codex login, a Claude login or API key, a paid Cursor plan, Grok Build, OpenCode, Antigravity. Use per-project settings to turn off providers a project should not use.

  2. Change the permission default

    Set new threads to Supervised or Auto-accept edits. Keep Full access for isolated machines. T3 Code says Auto uses automatic review on Codex, Claude, Cursor, Grok and fx and falls back to asking on OpenCode and Antigravity.

  3. Settle Claude billing

    T3 Code runs Claude through the Agent SDK. Anthropic's support page says Agent SDK and third-party app use draws from plan limits, and its legal page asks Agent SDK products to use API keys. Decide which your team uses and confirm it with Anthropic.

  4. Decide on telemetry and remote access

    Turn telemetry off with T3CODE_TELEMETRY_ENABLED=false if your policy requires it. Remote access works through T3 Connect, LAN pairing or Tailscale; T3 Connect needs a T3 account and allows 3 machines free.

What to verify before you buy

  1. Which permission mode new threads use on every engineer's install.
  2. Whether your Claude plans may run through the Agent SDK in T3 Code, or whether Claude runs on an API key.
  3. That telemetry is on or off by your policy, and which machines are linked through T3 Connect.
  4. That engineers install a pinned stable release and never a nightly build.

What to measure in your pilot

  • Whether Anthropic will state a position on Claude plan logins used through T3 Code's Agent SDK route.
  • How often stable releases change behavior that a team's settings depend on.
  • Whether T3 Tools will offer any organization admin, SSO or support terms.

Sources

Official documentation
16
Public source
3
  1. T3 CodeOfficial documentationt3.codes
  2. T3 Code READMEOfficial documentationgithub.com
  3. T3 Code license (MIT)Official documentationgithub.com
  4. T3 Code docs: FAQOfficial documentationgithub.com
  5. T3 Code docs: Install and first runOfficial documentationgithub.com
  6. T3 Code: DownloadOfficial documentationt3.codes
  7. T3 Code docs: Permission modesOfficial documentationgithub.com
  8. T3 Code docs: Thread sidebarOfficial documentationgithub.com
  9. T3 Code docs: Orchestrator MCP serverOfficial documentationgithub.com
  10. T3 Code docs: Remote accessOfficial documentationgithub.com
  11. T3 Code docs: Project settingsOfficial documentationgithub.com
  12. T3 Code docs: TelemetryOfficial documentationgithub.com
  13. T3 Code releases on GitHubOfficial documentationgithub.com
  14. T3 Code v0.1.6 release notes (Oct 31, 2026)Official documentationgithub.com
  15. T3 Code terms of serviceOfficial documentationt3.codes
  16. T3 Code privacy policyOfficial documentationt3.codes
  17. Anthropic: Claude Code legal and compliance (read Nov 1, 2026)Public sourcecode.claude.com
  18. Anthropic: Agent SDK overview (read Nov 1, 2026)Public sourcecode.claude.com
  19. Anthropic support: Use the Claude Agent SDK with your Claude plan (read Nov 1, 2026)Public sourcesupport.claude.com

Keep reading

  1. Buyer guideT3 Code Team Setup: Agents, Permissions and Remote AccessDecide which agents and accounts T3 Code may drive, set the permission mode, and lock down remote access before a team relies on it.
  2. ResearchSeveral Models in One App: Routing, Benchmarks and Native Tool Costs in T3 CodeWhich agent reaches which model in T3 Code, what independent benchmarks say about each kind of coding work, and when a model API's own search or video tool costs less than the data source's API.
  3. ReviewAider Review: Who Should Still Standardize on It in 2026Choose Aider for engineers who steer edits in the terminal on their own models, and pin a version, because releases have stalled.
Read the research