> ## Documentation Index
> Fetch the complete documentation index at: https://developers.t2000.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# create-t2-app

> Start a router-wired agent project in one command — npm create t2-app@latest.

Nothing to install first:

```bash theme={null}
npm create t2-app@latest
```

Interactive: **name → template → privacy mode**. Or scripted:

```bash theme={null}
npm create t2-app@latest my-app -- --template chat --privacy private --yes
```

## Templates

| id             | what you get                                                                                                                                                                                                                                        |
| -------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `agent-worker` | Headless TypeScript worker on `t2000/auto` — the smallest useful agent.                                                                                                                                                                             |
| `chat`         | Next.js streaming chat on `t2000/auto` — no AI SDK; the relay and SSE parser are two readable files.                                                                                                                                                |
| `sui-dapp`     | Sui dApp: wallet connect (dapp-kit), balance reads via gRPC (`SuiGrpcClient` — JSON-RPC retires the week of July 20, 2026), and an AI copilot that receives your holdings as context. Suggests the official Mysten `mystenlabs/skills` on scaffold. |

Every template is wired to the router out of the box (`api.t2000.ai/v1`,
model `t2000/auto`). First run: `npm install`, set `T2000_API_KEY` (free at
[agents.t2000.ai/manage](https://agents.t2000.ai/manage)), then `npm start`
(worker) or `npm run dev` (chat). The `x-t2000-served-model` response header
names the model that actually served each call.

## What's in a scaffold

Beyond the app itself, every template ships the agent layer:

* `AGENTS.md` — repo context any coding agent reads first.
* `plans/` — the plan-expensive / execute-cheap recipe (advisor writes
  `plans/<topic>.md`, a cheap executor like `t2code exec` runs it).
* `.t2000/config.json` — the per-repo privacy pin. `private` (default) keeps
  every call on open models with zero data retention; `full` allows frontier
  escalation; `confidential` is GPU-TEE only. [t2 code](/t2-code) honors the
  same pin, so the repo's privacy promise follows it to every contributor.

## Options

```
-t, --template <id>   agent-worker | chat | sui-dapp
    --privacy <mode>  private | full | confidential   (default: private)
-y, --yes             accept defaults for anything not provided
    --no-git          skip git init
```