Bivy

The open-source workspace for coding agents

Local coding agents that show their work. Anywhere.

Bivy runs Claude Code, Codex and other agents on your own machine. Reach them from a browser or your phone, with the app they’re building right in the chat. Try it, point at what’s wrong, and send it back.

Start free See how it works

10 free sessions a week · No credit card · Open source

A Bivy session: the request to move checked items into a Packed section, the agent’s reply, and a review card showing the running app before and after the change. Other sessions on Codex and Pi are listed alongside.
Real app · the agent’s change, shown as the running app

Works with any coding agent

01 Live previews

Don’t just read the diff. Try the app.

Bivy finds the dev server your agent starts and opens it next to the chat: web apps, terminals, even desktop apps. Here’s one change to a small demo app.

  1. Trail Pack open in Bivy’s preview drawer, with three checked items at the top of the packing list and the Point, Draw and Console controls below.

    01 · Preview

    Try the running app.

    Open it beside the chat. Here, packed items crowd the top of the list.

  2. Bivy’s Draw tool circles the checked items in Trail Pack. A note asks the agent to move them into a Packed section while keeping selections and the count.

    02 · Feedback

    Circle it. Say what you want.

    Your drawing and note go to the agent with the page context. No screenshots to paste.

  3. Back in the Bivy chat, the agent’s reply is followed by a review card showing Trail Pack with the new Packed section, and a Before / Now switch.

    03 · Review

    See before and now.

    The agent’s reply comes with the app as it looks now. Flip to Before, open it live, or share it.

02 Automations

Work starts without you. It comes back ready to try.

Send a GitHub issue, a failed CI run, a Linear ticket, a Slack message, a schedule or any webhook to Bivy. An agent picks it up on your machine and posts back a preview link or a pull request.

01 / The trigger

A task arrives.

“Build a weekend packing checklist.”

Here, a Basecamp to-do fires a webhook.

Webhook

02 / On your machine

Your agent builds it.

It writes the code and runs your tests in your repo, with your tools.

Preview link

03 / Back to you

You try it where the task lives.

The preview link lands in the to-do. Open it, try the app, and reply in the same session.

“Separate packed items from what is left.”

Your feedback · same session
You review every change and decide when to merge.

Start work from

  • GitHub issues
  • Failed CI
  • Linear
  • Slack
  • Basecamp
  • Schedules
  • Signed webhooks
Automation recipes

03 Any agent

Use every agent you pay for. Switch mid‑task.

Hit a usage limit? Keep going.
Fork the session, with its code and conversation, to another agent. Or let Bivy pick it up again when the limit resets.
Pick the right agent for the job.
Claude Code on your Mac, Codex on a Linux server, side by side in one list.
Get a second opinion.
Ask one agent to review another’s work, on the same machine or a different one.
Log in once.
Keys and logins sync across your machines, end-to-end encrypted.
A Claude Code session in Bivy that hit its usage limit, offering Fork to another agent and Retry automatically when the limit resets.
Real app · a usage limit, two ways forward

04 Your machine, anywhere

Runs where your code already lives.

Read the security model

Free to start. Simple after that.

Free

$0

  • 10 new sessions every week, including automations
  • 5 hours on a Bivy computer, once
  • Live previews, feedback and share links
  • All integrations included
  • No credit card
Start free

Cloud

$15/mo

  • Unlimited sessions, including automations
  • 120 hours a month on one Bivy computer, then $0.10/hour
  • Remove “Made with Bivy” from shared previews
  • Support from the Bivy team
  • Cancel anytime
Start free, upgrade in the app

Self-host

$0

  • The whole stack, open source under AGPL-3.0
  • Your infrastructure, your operations
Self-host guide

Your own computers are unlimited: you bring the machine, and sessions you start on it never count. A Bivy computer sleeps when the work stops, and sleeping is free. Extra hours are $0.10, up to a ceiling you set. You bring your own agent subscription or API key.

Your first session in three steps.

  1. Sign in at app.bivy.sh.
  2. Connect a machine. Paste the one-line install command into a terminal in your repo.
  3. Start a session with the agent you already use.
Start free

Questions.

What is Bivy?

A workspace for your coding agents: chat, terminal, live app previews, automations and code review, from your desk, browser or phone. It’s not another coding agent. Your own machines are the normal place work runs. A Bivy computer is optional, and it sleeps when you stop.

Does my machine need to stay on?

Your own machine has to stay awake while it runs the work. Use an always-on server if you want that work to continue while your laptop is closed. A Bivy computer sleeps when the work stops and wakes when you come back.

Which agents can I use?

19 agents are supported, including Claude Code, Codex, Pi, OpenCode, Gemini CLI, Grok, Cursor, GitHub Copilot, Amp, Goose and Aider. Any other CLI or ACP agent can be connected through a generic adapter (experimental). Capabilities such as resume, model choice and approvals vary by agent.

All agents and what each supports ↗
What carries over when I switch agents?

A fork or move takes the code and conversation context with it. How much history transfers depends on the agents involved, so check the new session before retiring the original. Switching doesn’t reset the first provider’s limit, and automatic retry needs a reset time from the provider.

How session handoffs work ↗
What do automations need?

An automation rule for the source (GitHub, Linear, Slack, a schedule or a signed webhook). To post a preview link back to a tool like Basecamp, the agent needs credentials for that tool.

Webhook recipes ↗
Cloud or self-hosted?

Cloud hosts the app and relay. Work runs on your own machine, or on an optional Bivy computer that sleeps when you stop. You still supply model access. Self-host the AGPL-3.0 software for $0 and run your own infrastructure.

Self-host guide ↗
What is encrypted, and what isn’t?

Interactive prompts, transcripts, diffs, files, and model keys are end-to-end encrypted between your machine and paired devices. Your model provider receives what your agent sends; Slack commands and generic webhook instructions reach the routing service as plaintext. Anyone with a preview share link can use that app until the link expires or you revoke it.

Security boundaries ↗
Are agents sandboxed?

Not universally. Some agents have native sandboxing or structured approvals; others run largely as you. Choose permissions and isolation appropriate to the agent and task.

Agent security model ↗