Supported providers

The four coding agents T3 Code supports, how to authenticate each one, which is best tested, and what to do about Gemini, Ollama and other agents people ask for.

Last reviewed 2026-07-27 · verified against the pingdotgg/t3code repo

T3 Code drives coding agents; it doesn't replace them. Four are supported today, and you install and authenticate each one yourself.

The four#

Provider	Auth command	Maturity in T3 Code
Codex	`codex login`	Best tested — the original integration
Claude Code	`claude auth login`	Well supported, actively developed
Cursor	`cursor-agent login`	Supported via the agent CLI
OpenCode	`opencode auth login`	Supported, widest model choice

More are described as coming. What about Gemini, Ollama and the rest? →

Setting one up#

Same three steps regardless of provider:

bash

# 1. install the CLI
npm i -g @openai/codex

# 2. authenticate
codex login

# 3. prove it works on its own
codex --version

Only then start T3 Code. If step 3 fails, T3 Code cannot help you — it launches that exact binary.

Bring your own key#

There's no T3 Code subscription and no T3 Code API key. You use the account you already have:

An OpenAI subscription or API key, via Codex
An Anthropic subscription or API key, via Claude Code
A Cursor subscription, via the Cursor agent CLI
Whatever OpenCode is configured against

You pay your provider directly. T3 Code adds nothing to the bill.

Choosing between them#

Genuinely: use whichever one you already pay for. The differences in T3 Code are smaller than the differences between the underlying agents, and you'll get more from a subscription you already have than from optimising the wrapper.

If you have several, some rough guidance:

Codex — the deepest integration and the one the maintainers use most. Fewest surprises.
Claude Code — strong on multi-file reasoning; recent builds surface Claude Code skills directly in the composer and added Claude Opus 5 support.
Cursor — sensible if you're already in the Cursor ecosystem and want its agent outside the IDE.
OpenCode — the open one, and the most flexible about which model sits behind it.

Switching mid-thread#

You can move between models and agents within a workspace thread rather than managing separate terminal sessions. That's more useful than it sounds: cheap fast model for exploration and boilerplate, expensive careful model for the part that matters, without losing thread context.

When a provider breaks#

Almost all provider problems are version skew or PATH, not T3 Code:

Update the CLI. npm i -g @openai/codex@latest. This fixes the plurality of issues.
Check PATH. Works in your terminal but not in T3 Code? PATH inheritance.
Re-authenticate. Sessions expire. codex login again.
Check your provider account. Rate limits and quota exhaustion surface as confusing mid-run errors.

Full troubleshooting index →

FAQ#

What providers does T3 Code support?#

Codex, Claude Code, Cursor and OpenCode, with more described as coming. Each requires its own CLI installed and authenticated on the machine running the T3 Code server.

Do I need a T3 Code subscription?#

No. T3 Code is free and open source under the MIT license. You bring your own provider account — an OpenAI, Anthropic or Cursor subscription, or an API key — and pay that provider directly.

Can I use more than one provider in T3 Code?#

Yes. You can authenticate several CLIs and switch between models and agents within a workspace thread rather than running separate terminal sessions.

Which T3 Code provider works best?#

Codex has the most mature integration since it was the original target. Claude Code is a close second and actively developed. In practice, use whichever agent you already have a subscription for.