context-labs/whip
A fast coding-agent harness in Go. Tool-use loop, bubbletea TUI, provider-routable models with live catalog discovery, MCP support, background subagents. One binary, no runtime, built for open-source models.
whip – a fast coding‑agent harness written in Go
What it is – whip is a single‑binary command‑line tool that runs an LLM‑driven “tool‑use” loop. It lets a language model read, write, edit files, run bash commands, and even spawn sub‑agents, all from an interactive terminal UI built with Bubble Tea.
Why it exists – The author argues that existing agent harnesses are slow or require a lot of configuration. By using Go’s concurrency primitives (channels, goroutines) the program can make parallel tool calls, stream responses, and keep the path from the model to the user as short as possible.
Key features (as described in the README)
- Speed‑focused design – parallel tool calls, streaming I/O, no extra runtime layers.
- Opinionated defaults – pulls together ideas from projects like pi, opencode, codex, and exo; the roadmap file cites each source.
- Provider‑routable models – works with any OpenAI‑compatible endpoint. It can auto‑discover every model a provider offers (e.g., OpenRouter) without per‑model configuration.
- Built‑in model discovery –
whip auth openrouterloads the whole OpenRouter catalog;whip auth codexlets you log in with a ChatGPT/Codex subscription instead of an API key. - Automatic token handling – login tokens are stored locally (
~/.codex/auth.json), refreshed silently, and never exposed in prompts. - MCP support – dropping a
.mcp.jsonfile in a repository makes “MCP servers” appear, accessible via the/mcpcommand. - Theming –
/themelets you preview and select terminal colour schemes. - Convenient commands –
/context‑doctor– audits token usage of a fresh session./goal <text>– tells the agent to work until the goal is achieved./model– interactively pick a model, with live updates from the provider catalog./usage– shows rate‑limit information for a Codex subscription.
Installation
# One‑liner installer (downloads a checksum‑verified binary for Linux/macOS)
curl -fsSL https://raw.githubusercontent.com/context-labs/whip/main/install.sh | sh
Or from source (requires Go ≥ 1.27):
go install github.com/context-labs/whip/cmd/whip@latest
After installation the whip command is ready to run.
Updating – whip update re‑runs the installer script to fetch the newest release.
Documentation – The repository contains a docs/ folder with detailed markdown files covering architecture, the agent loop, concurrency model, full feature map, and a roadmap that distinguishes shipped features from upcoming work.
Typical first steps
/context-doctor # see token injection details
/goal Build a simple CLI tool
/model # pick a model (type to filter)
/theme # choose a terminal theme
Press Ctrl‑C once to interrupt the current operation, twice to exit the program.
All information above is taken directly from the repository’s README; no additional assumptions have been made.
Related
- Project
- Project
- Project
- Project