bud.sh

Built for your agents,
not instead of them.

Bud doesn’t compete with your editor, terminal or coding agent. It’s the layer between you and them: where you look at the result, point at what’s next and hand over precise context. And when you want, it runs an agent on a task itself, in a sandbox, reviewed before anything is pushed.

Working with agentsSee the workflow

Whatever runs your code,
Bud fits next to it.

Terminal agents, editors, desktop apps or your own scripts. Bud needs no plugin for them, because everything it keeps is plain files and Git.

  • Claude Code
  • Codex
  • Cursor
  • Gemini CLI
  • GitHub Copilot
  • OpenCode
  • Amp
  • Windsurf
  • Cline
  • Zed
  • Kiro
  • Warp
  • Claude
  • ChatGPT
  • Factory
  • Kilo Code
  • OpenHands
  • Qwen Code
  • Antigravity
  • Crush
  • OpenClaw
  • MCP agents

A layer,
not another harness.

Your agent works in your repository like it always has. Bud sits between you and that work: it shows the result, keeps the conversation and turns it into context the next prompt can use, for your agent or for one Bud runs.

Hand off a change
YouLook, decide and point
comments and context
BudCanvas · Tasks · Agents · Files · Comments
a precise handoff
Your agentWrites the code
edits and commits
Your repositoryCode, comments and tasks in Git

Each does
what it’s best at.

Bud

Where people look, decide and agree.

  • Shows the resultEvery page live on one canvas, next to the code.
  • Keeps the conversationPins, replies and statuses, saved to Git.
  • Holds the planA board for the issues your agent works on.
  • Keeps you in chargeAgents it runs push only what you approve.

Your agent

Where the code gets written.

  • Reads and edits codeIn your repository, with your tools.
  • Runs your commandsTests, builds and the dev server you already use.
  • Commits the changeThrough your usual Git workflow.
  • Or works in a sandboxStarted by Bud on a task, on its own branch.

From a detail
to a finished change.

Hand the change to your own agent, or start one on the task in Bud. Bud never starts a run on its own: you stay in the loop, and every step leaves a record.

  1. Point

    Pin the detail on the page or select the lines.

  2. Agree

    Reply until the change is clear, then mark it in progress.

  3. Hand off

    Copy the thread into your agent, or make it a task and start one in a sandbox.

  4. Check

    Review the preview or the canvas, then approve the push or resolve the thread.

Or let Bud
run one for you.

Make a task from a comment and start it. A coding agent, OpenCode by default, works on it in a sandbox with its own dev server: a worktree on your machine or a fresh one on boat.dev. Follow it in the chat or its terminal, ask for changes, then approve. Only then does Bud push, and only the task’s own branch.

Explore agents

Your agent already
speaks Bud’s language.

Tasks and comments live in Beads, an issue tracker built for agents, and agents Bud runs read them through its read-only task tools. The docs come as Markdown. Nothing needs an integration to be read.

The Beads integration

Terminal

# What the team discussed
bd list --type message --include-infra

# What's ready to work on
bd ready

# How Bud works, for your agent
curl https://bud.sh/llms.txt

Give your next change
a little Bud.

Add Bud to Vite, pick a few pages and start the conversation. Free, and one npm install away.

Read the setup guide