t3chat-cli

A command-line interface for t3.chat

Prerequisites

• A t3.chat account
• curl

Installation

bun add -g @cyanlabs/t3chat

Once installed, the CLI is available as the t3chat command.

Setup

1. Authenticate

t3chat auth login

This opens t3.chat in your browser and walks you through extracting session cookies:

1. Log in to t3.chat
2. Open DevTools (Cmd+Option+I)
3. Go to the Network tab
4. Send any chat message
5. Click the /api/chat request
6. Copy the Cookie header value
7. Paste it into the CLI prompt

Your credentials are saved to ~/.config/t3chat-cli/config.json.

2. Select a model (optional)

t3chat models

Fetches the full model catalog from t3.chat and presents an interactive picker. Models are displayed with their provider and cost tier:

Claude 4 Sonnet (Anthropic) $$$
GPT-4.1 (OpenAI) $$
Gemini 2.5 Pro (Google) $$
DeepSeek R1 (DeepSeek) $
Kimi K2.5 (Moonshot) $
...

The default model is kimi-k2.5 if none is selected.

3. Install agent skills (optional)

t3chat skill

Generates and installs agent skill files for detected AI coding agents (Claude Code, OpenCode). This lets your coding agent understand and use t3chat commands. An interactive prompt lets you toggle which agents to install/uninstall skills for.

Skills are also auto-updated silently when a new CLI version is detected.

Usage

t3chat "What is the capital of France?"

t3chat explain the difference between TCP and UDP

The prompt is variadic -- quotes are optional for multi-word inputs. Responses stream directly to stdout.

After the response, an interactive prompt appears so you can send follow-up messages with full conversation context. Press Enter on an empty line or Ctrl+C to exit.

Options

| Flag | Alias | Type | Default | Description | | ----------------- | ----- | ------- | ------- | ---------------------------------------------------------------- | | --model | -m | string | | Override the default model for a single query | | --search | -s | boolean | false | Enable web search | | --searchLimit | -S | number | 1 | Number of searches the agent is allowed to make per response | | --interactive | -i | boolean | true | Enable interactive follow-up prompts (--no-interactive to disable) | | --version | -v | | | Show version number | | --help | -h | | | Show help message |

Configuration

All configuration is stored in ~/.config/t3chat-cli/config.json:

{
  "cookies": "convex-session-id=...; ...",
  "model": "claude-4-sonnet"
}

| Field | Description | | --------- | -------------------------------------------------------- | | cookies | Session cookies from t3.chat (set via t3chat auth login) | | model | Selected model ID (set via t3chat models) |

Development

# Run in dev mode (executes TypeScript directly)
bun run dev

# Type-check
bun run check:types

# Build standalone executable
bun run build

# Run the built binary
bun run start

Built With

• Bun -- Runtime and package manager
• Crust -- CLI framework (@crustjs/core, @crustjs/plugins, @crustjs/prompts)
• TypeScript