btw

AI answers in your terminal.

https://github.com/user-attachments/assets/47880a03-6207-4b6c-a8bb-7f781988cec1

Asks a Question → Gets an Answer (with optional Thinking). That's it.

Note: btw, qq, and q all run the same tool — use whichever is shortest.

Install

npm install -g @notyato/btw
bun add -g @notyato/btw

Quick start

# Ask anything - no setup required
btw "what is 2+2"
btw "how do I update a dep to latest with bun"

# Add provider
btw connect

# Select model
btw model

Sessions are automatic — every question builds on the previous one in a shared global context.

Usage

btw <question>            Ask a question (inline mode)
btw                       Open multiline input (interactive mode)
btw connect               Store an API key for a provider
btw model                 Set your default model
btw shell                 Print export BTW_SESSION_ID for per-terminal sessions
btw shell --install       Add session init to your shell config
btw session global        Switch to global session mode
btw session per-terminal  Switch to per-terminal session mode
btw --help                Show help
btw --version             Print version

Options

| Flag | Description | |------|-------------| | --no-thinking | Hide thinking/reasoning output | | --model <provider:model> | Override the model for this question |

Model override

Run any question with a different model, no reconfiguration needed:

btw --model opencode:deepseek-v4-flash-free "explain quantum computing"

Sessions

btw keeps a conversation history so each question sees prior context.

| Mode | Behavior | |------|----------| | Global (default) | All questions share one session stored in ~/.config/btw/sessions/GLOBAL.json. Run btw session global to switch here. | | Per-terminal | Each terminal gets its own session identified by BTW_SESSION_ID. Run btw session per-terminal to enable, then add eval $(btw shell) to your shell config (or use btw shell --install). |

Session errors never interrupt your question — warnings are logged and btw continues without context.

Setup

To use a different provider and model:

1. btw connect — pick a provider and enter your API key. This stores a credential and makes that provider's models available.
2. btw model — choose a default model from your connected providers.

Configuration

| Data | Path | |------|------| | Config (model, preferences) | ~/.config/btw/config.json | | Credentials (API keys) | ~/.cache/btw/auth.json | | Session data | ~/.config/btw/sessions/GLOBAL.json | | Per-terminal sessions | ~/.cache/btw/sessions/{ulid}.json |

Features

• Streaming Markdown renderer — headings, code blocks with syntax highlighting, tables, blockquotes, lists, links (OSC-8 hyperlinks), and more
• Thinking/Reasoning support — model chain-of-thought rendered in dim text on stderr (--no-thinking to hide)
• Inline & ~~Interactive modes~~ — pass a question as an argument or open a multiline prompt
• Custom model override — --model provider:id for one-shot model switching without reconfiguration
• Conversation sessions — automatic global context by default, per-terminal opt-in via btw session

License

MIT — see LICENSE.