taui 0.3

Agentic coding interface you can reshape

taui

Agentic coding interface you can reshape.

Alpha: This project is still in early development. APIs, commands, and behavior may change without notice.

Install

uvx taui

Or install permanently:

uv pip install taui

What is taui?

Taui is a highly customizable agentic coding interface. Instead of adapting your workflow to a fixed assistant, you control the interface itself: UI, agent, tools, prompts, extensions, skills, and storage.

Taui is a full-screen Textual TUI. Running taui launches the terminal interface with a sidebar, scrollable chat history, live streaming, visual tool status, approvals, questions, steering, and queueing.

Requirements

• Python 3.13+
• A supported LLM provider account (see below)

Getting Started

First Run

When you run taui for the first time, it launches an interactive provider selection prompt. Pick one or both providers and follow the auth flow:

taui

You can also authenticate explicitly at any time:

taui --login

Providers

Taui currently ships with two built-in providers:

Provider	Auth Method	What You Need
GitHub Copilot	OAuth device flow	A GitHub account with an active Copilot subscription
OpenAI Codex	PKCE browser redirect	A ChatGPT Plus or Pro account

GitHub Copilot

1. Run taui or taui --login
2. Select GitHub Copilot
3. A device code is displayed — open the GitHub URL shown and enter the code
4. Once authorized, your token is saved to ~/.config/taui/config.toml
5. Subsequent runs use the saved token automatically

OpenAI Codex (ChatGPT Plus/Pro)

1. Run taui or taui --login
2. Select OpenAI Codex
3. A browser window opens for OpenAI OAuth — sign in and authorize
4. The token is saved to ~/.config/taui/config.toml

Choosing a Provider and Model

# Start with a specific provider
taui -p copilot
taui -p codex

# Use a specific model
taui -p copilot -m claude-sonnet-4

# Work in a specific directory
taui -d /path/to/project

# Resume a previous session
taui --session <session_id>

# Check version
taui --version

Within the TUI, use /model to switch models and /provider to switch providers.

Tool Approval Policy

By default, destructive tools (bash, write, edit) require user confirmation before executing. Read-only tools (read, glob, grep) run automatically.

You can customize this in your config file (~/.config/taui/config.toml):

[taui.tool_policy]
bash = "auto"       # skip confirmation for bash
write = "confirm"   # require confirmation (default)
edit = "deny"       # block entirely

Valid policy values: auto, confirm, deny.

Approval prompts also support persistent auto-approval for a whole tool. Choosing the project option writes a generated extension to .taui/extensions/; choosing the global option writes it to ~/.taui/extensions/. The generated extension replaces the tool with an equivalent wrapper and sets that tool's policy to auto when extensions load. Project scope is the default persistent choice.

Key Bindings

Key	Action
Ctrl+Q	Quit
Ctrl+N	New session
Ctrl+C	Cancel active request or approval
Ctrl+B	Toggle sidebar
Ctrl+X	Context breakdown
Escape	Leave self-edit mode

Slash Commands

Command	Description
/help, /h, /?	Show help
/model	Switch model
/provider	Switch provider
/compact	Compact conversation (shows token savings)
/clear	Clear chat
/cost	Show session cost
/sessions	Browse and resume sessions
/new	Start new session
/reload	Hot-reload extensions
/extensions	List extensions
/i	Enter self-edit mode
/login	Re-authenticate providers
/logout	Clear saved credentials
/copy	Copy last response
/export	Export session
/hotkeys, /keys	Show key bindings
/verbose, /quiet	Toggle tool output verbosity

Extending Taui

Extensions

Extensions are Python files that register tools, commands, hooks, and skills.

Locations:

• Global: ~/.taui/extensions/*.py
• Project: .taui/extensions/*.py

Example extension:

def register(ctx):
    ctx.tools.register(my_tool)
    if ctx.commands:
        ctx.commands.register(my_command)
    if ctx.hooks:
        ctx.hooks.add("system_prompt", my_transform)
    ctx.skills.add_path("skills/my-skill.md")

Use /reload to hot-reload extensions without restarting. Errors from individual extensions are reported in the chat log.

Skills

Skills provide specialized instructions loaded lazily into the agent context.

Locations:

• ~/.config/agents/skills/<name>/SKILL.md
• ~/.taui/skills/<name>/SKILL.md
• .agents/skills/<name>/SKILL.md
• .taui/skills/<name>/SKILL.md

Self-Edit Mode

Run /i to enter self-edit mode — a specialist agent that can create or modify extensions, skills, commands, and tools through the extension surface.

Configuration

Config is loaded from ~/.config/taui/config.toml with this structure:

[taui]
provider = "copilot"
model = "claude-sonnet-4"
max_turns = 50
verbose_tools = true

[taui.tool_policy]
bash = "confirm"
write = "confirm"
edit = "confirm"

CLI arguments and environment variables override config file values.

Data Storage

Taui stores session data in .taui/store.db (SQLite, WAL mode) in the working directory. Sessions are append-only event streams that support replay.

Credentials are stored at ~/.config/taui/config.toml. Logs are at ~/.taui/.logs.

License

MIT