Gemini, Ollama and other agents
An honest status check on Gemini, GitHub Copilot, Ollama, OpenRouter and local models in T3 Code — and the workaround that actually gets you there today.
These are among the most-searched T3 Code questions, and they deserve a direct answer rather than a hopeful one.
The supported list, as of this review#
Four providers are documented as supported: Codex, Claude Code, Cursor and OpenCode. The project says more are coming, but anything not on that list is not a first-class provider today.
So the direct answers:
| Asked for | First-class provider? |
|---|---|
| Gemini / Gemini CLI | Not on the supported list |
| GitHub Copilot | Not on the supported list |
| Ollama / local models | Not directly — see below |
| OpenRouter | Not directly — see below |
| Grok, DeepSeek, other models | Not directly — see below |
Why "model" and "provider" aren't the same question#
Most of these questions conflate two layers, and separating them makes the answer obvious.
T3 Code → provider CLI → model backend
(UI, git) (agent logic) (the actual model)T3 Code integrates at the provider CLI layer. It has no model integration at all — it doesn't know what a model is, only how to talk to an agent CLI over JSON-RPC.
So "does T3 Code support Gemini?" really means one of two things:
- Is there a Gemini agent CLI adapter? Not on the supported list today.
- Can I run a Gemini model through T3 Code? Yes, if a supported provider can be pointed at it.
The workaround: OpenCode#
For nearly every "can I use model X" question, OpenCode is the answer.
It's an agent designed to work with many model backends rather than one vendor's. Configure OpenCode against the model you want — including local ones — and T3 Code drives OpenCode. You get the whole T3 Code workflow (worktrees, diffs, one-click PRs, remote access) on top of whatever backend OpenCode supports.
# configure the model in OpenCode first, then:
opencode auth login
npx t3@latestThe key point: configure the model in OpenCode, not in T3 Code. T3 Code's picker reflects what the provider exposes.
Local models specifically#
If you're chasing Ollama or a self-hosted endpoint, the question to answer isn't about T3 Code — it's whether OpenCode can be configured against your local endpoint. Get that working in OpenCode alone first. Once opencode runs against your local model in a terminal, T3 Code will drive it.
Be realistic about capability, though. Agentic coding is one of the harder workloads: long contexts, reliable tool calls, many turns. Smaller local models often do fine at chat and struggle badly at autonomous multi-file editing. Test on a real task before committing to the setup.
If you want a provider added#
It's an MIT-licensed project that reached four providers quickly, which suggests the adapter surface is tractable. If you want a fifth:
- Check the issues for an existing request and add to it
- Ask in Discord, where maintainers are active
- Read an existing adapter and open a PR — that's the path that actually ships
FAQ#
Does T3 Code support Gemini?#
Gemini is not on the documented list of supported providers, which currently covers Codex, Claude Code, Cursor and OpenCode. Since the project ships nightly, check the official docs and repository for current status.
Can I use Ollama or local models with T3 Code?#
Not directly. T3 Code integrates with agent CLIs, not models. Configure OpenCode against your local endpoint and T3 Code will drive it, provided the model is capable enough for agentic work.
Does T3 Code work with GitHub Copilot?#
Copilot is not on the documented supported provider list. T3 Code currently supports Codex, Claude Code, Cursor and OpenCode.
Can T3 Code use OpenRouter?#
Not as a direct integration. Route it through a provider that supports OpenRouter as a backend — OpenCode is the most flexible option — and configure it there rather than in T3 Code.
Why doesn't T3 Code just support every model?#
Because it isn't a model client. It speaks JSON-RPC to agent CLIs, which handle model access themselves. Adding a model means either the provider supports it, or someone writes a new provider adapter.