CODING AGENTS ON MACHINES YOU CONTROL

Run coding agents on your machines. Control them from anywhere.

Run Claude Code, Codex, Pi, OpenCode, and other coding agents on machines you control. Start sessions yourself or from GitHub, Slack, CI, webhooks, and schedules — then inspect, steer, approve, and continue them from your phone, browser, or terminal.

$ curl -fsSL https://bivy.sh/install.sh | bash

See automations → Read the quickstart → View source →

macOS or Linux · open source (AGPL-3.0) · unlimited machines · bring your own agents, subscriptions, and model accounts · free to self-host, or try Bivy Cloud free

A Bivy agent session on a phone showing a finished coding task, opened pull request, and composer to continue. — Start it manually or automatically. Join the same session from terminal, browser, or phone.

Hero demo video placeholder

Suggested 12–18s loop: bivy run claude in terminal → agent keeps working → phone notification → approve / reply → PR opens. This should sit directly under the hero on launch.

Terminal

Start it in your terminal

Keep the CLI beside your editor. One command starts a session — pick the machine, agent, and model, then hand it off to any device.

$ bivy run claude $ bivy run codex $ bivy run pi --model gpt-5.6-sol

Terminal

Phone / browser

Start it remotely

Open a session from your phone or browser and choose the machine, agent, model — even the GitHub repo — before it starts running.

machinehome-lab-8 agentclaude modelgpt-5.6-sol repobivysh/bivy

Phone / PWABrowserVoice

Automated

Or let events start it

GitHub, Slack, Linear, CI, schedules, and webhooks can open the same kind of session — routed to the machine, agent, and model you choose.

GitHubSlackLinearCIScheduleWebhook

→ machine→ agent→ model

Either way, it becomes the same Bivy session. Join it from terminal, browser, or phone. Same agent, transcript, branch, and worktree.

Bring your agents

Claude Code Pi Codex OpenCode and more: Beta Use provider sign-in, subscriptions, or your own API keys. Agent matrix →

Automations

Automations start real sessions — not background jobs.

GitHub, Slack, Linear, CI, schedules, and webhooks can start the same kind of Bivy session you launch from the terminal. Route each one to the right machine, agent, and model, then join it whenever human judgment is needed.

01 · TRIGGER

Something happens

A label, CI failure, @mention, schedule tick, or signed webhook arrives.

02 · TEMPLATE

The job is defined

Prompt, policy, and success criteria are already chosen.

03 · ROUTE

Machine · agent · model

Use defaults, repo preferences, or an explicit event override.

04 · SESSION

You can join it

Inspect progress, approve tools, reply, or take over live.

Triggers

Front doors into a session — not a separate job system.

GitHub — labels, @mentions, CI failuresissue · check · comment
Slack — @bivy and slash commandschannel · default repo
Linear — assign or mentionteam · label
Schedule — cron in your timezonerepo required
Signed webhook — CI, Sentry, alerts, custompayload → prompt

Route each session

Unattended work still lands where the tools, repository, and secrets live.

Machine

Default node, repo default, or explicit tag — bivy/macbook, on hetzner.

Agent

Claude Code, Codex, Pi, OpenCode, and more.

Model

Pin a model on the automation or pass it on the event. Your subscriptions and keys stay yours.

Automation demo video / GIF placeholder

Suggested 10–15s loop: GitHub issue or CI failure → Bivy opens session on selected machine → agent starts → phone opens same session → human adds one instruction → PR updates.

Persistent sessions

Start anywhere. Continue the same session anywhere.

Start from the CLI, a trigger, or the app. Leave it running on the machine where it started, then continue from your phone, browser, or terminal without rebuilding context.

01 · START

Launch the agent

Choose a machine, repository, agent, and model — manually or from an automation.

02 · LEAVE

Let it keep working

Walk away while the session continues on the machine where it started.

03 · CONTINUE

Open it as chat

Read progress, reply, speak, attach files, and approve tools from your phone.

04 · RETURN

Pick up anywhere

Return to terminal or browser, or fork the session to another agent or machine.

Bivy on a phone listing agent sessions across machines, with PR badges. — **A real chat app, not a terminal viewport.** Sessions, diffs, approvals, and follow-ups in one mobile interface.

Bivy on a phone showing a file diff the agent just made. — **A real chat app, not a terminal viewport.** Sessions, diffs, approvals, and follow-ups in one mobile interface.

Mobile session GIF placeholder

Suggested 8–12s loop: session list → open run → inspect diff → approve tool → send follow-up. Use this as a tighter alternative to three static phone screenshots if desired.

The right interface for the device

Terminal when you want it. Chat when you don't.

Bivy does not replace the terminal. It gives the same coding-agent session a proper app interface when a terminal is the wrong tool — especially on a phone.

Talk naturally. Send text, voice, and attachments.
See what needs you. Questions, approvals, failures, and finished runs land in one inbox.
Inspect and continue. Review files, diffs, checks, branches, pull requests, time, and cost.
Drop into a real PTY. Use the terminal whenever chat is not enough.

The Bivy desktop app with the session list and a live coding-agent conversation.

Ownership

Your machines. Your agents. Your accounts. Your keys.

Bivy connects the infrastructure and subscriptions you already own. It does not sell you compute or model access.

Your machines

Run agents on your laptop, Mac mini, Linux server, home lab, or cloud VM. Repositories, tools, network access, and compute stay in your environment.

Your accounts and keys

Use Claude sign-in, Codex sign-in, supported OAuth, or your own API keys. Bivy does not provide or resell models.

Your fleet

Route each session to the machine and agent that fit the job, while keeping approvals and supported credentials available across the fleet.

 GitHub · Slack · Linear · Webhooks · Schedules
                              │
                              ▼
                   ┌─────────────────────┐
                   │      Bivy app       │
                   │ persistent sessions │
                   │ manual + automated  │
                   └─────┬─────┬─────┬───┘
                         │ E2E │
             ┌───────────┘     │     └───────────┐
             ▼                 ▼                 ▼
   ┌─────────────────┐ ┌─────────────────┐ ┌─────────────────┐
   │  Work laptop    │ │  Home server    │ │    Cloud VM     │
   │ $ bivy run      │ │ $ bivy run      │ │ $ bivy run pi   │
   │   claude        │ │   codex         │ │   --model       │
   │                 │ │                 │ │   gpt-5.6-sol   │
   └─────────────────┘ └─────────────────┘ └─────────────────┘

 Provider sign-ins and API keys belong to you.
 Agents and code run on your infrastructure.

Bivy is the session and coordination layer. Agents and code run on your infrastructure; Bivy connects, routes, and exposes them through terminal, browser, and phone.

Security and ownership

Bivy coordinates the work. It does not become the model provider.

Your agents run on your machines with your provider accounts, subscriptions, and credentials. Bivy-managed credentials sync end-to-end encrypted between your devices and machines.

Security model →

The boundary matters. Your model provider still receives whatever the agent sends to it. Bivy Cloud handles account, routing, billing, and run-status metadata.

Machines and computeYours

Claude, Codex, and model subscriptionsYours

API keys and supported OAuth credentialsYours; E2E-encrypted sync

Repositories, tools, and agent processesRun on your machines

Prompts and transcripts in remote accessE2E encrypted to your devices

Inference billingPaid directly to your provider

Features

Routing, forking, and rules.

Beyond start-and-continue, Bivy gives you precise control over where a session runs, where it goes next, and what happens when work fails — for manual runs and automations alike.

Routing

Send every session to the machine, agent, and model that fit the job. Set defaults per automation or per repository, override per event, and let offline tagged hosts wait instead of being silently rerouted.

Forking

Fork a session onto another agent or machine — or move it outright — keeping the transcript and any uncommitted worktree changes. Switching agents replays the conversation or seeds a continuation.

Rules

Decide how unattended work behaves: routing rules that place each event, plus retry and fallback behaviour for failed or stalled runs — applied the same whether the session started by hand or from a trigger.

Compare

Why not just SSH, tmux, or a hosted coding agent?

SSH gives you a terminal. Hosted agents give you someone else's infrastructure. Bivy keeps the agents on your machines while adding persistent sessions, mobile-native interaction, approvals, routing, and automation.

Comparing Bivy, phone terminal or SSH, hosted agents, and IDE agents.
What you need	Bivy	Phone terminal / SSH	Hosted agent	IDE agent
Mobile experience	Purpose-built chat app	Desktop terminal on a phone	Job dashboard or vendor UI	Usually desktop-first
Terminal access	Yes, alongside chat	Primary interface	Usually limited	Inside the IDE
Where agents run	Your machines	Your machines	Vendor infrastructure	Your workstation
Model accounts	Your subscriptions and keys	Your subscriptions and keys	Vendor plan or catalog	Usually vendor-linked
Session continuity	Phone, browser, terminal, machines, agents	Manual terminal state	Within the hosted run	Within the editor
Automation	Triggers start the same joinable session	Build the glue yourself	Often a separate job UI	Rarely central

Use Bivy when you want coding agents on your own infrastructure, but don't want the terminal to be the only place you can control them.

Pricing

Open source and free to self-host. Cloud is the same product, hosted.

Every feature is free when you self-host. Cloud gives you the same product on Bivy Cloud so there is no relay or control plane for you to operate.

Open source

Self-hosted

Run the whole stack on your own infrastructure.

Open source under AGPL-3.0
Every feature, unlimited
Unlimited interactive sessions
Automations and routing
Unlimited machines and paired devices
Community support

Install free

Hosted & managed

Cloud

$15/mo

The same Bivy, fully managed.

25-session free trial — no card
Everything in Self-hosted, hosted for you
Managed relay + app
Automatic upgrades, backups, and TLS
Unlimited sessions once subscribed
Email support

Start free trial

Design partners

Team

Let's talk

Early team features built with design partners.

Everything in Cloud
Shared routing and approvals
Onboarding and deployment review
Shape roles, SSO, and audit export
Founding-customer pricing
Priority support

Become a design partner

FAQ

Common questions

Bivy is 0.x. We use it every day, but things still move, and not every agent adapter is equally mature.

What is Bivy?

Bivy runs coding agents on machines you control as persistent sessions. You can start a session yourself or let GitHub, Slack, Linear, CI, webhooks, or schedules start one automatically, then join and steer it from terminal, browser, or phone.

How do automations relate to sessions?

An automation starts an ordinary Bivy session. You can inspect it, message it, approve tools, and take it over exactly like a session you started by hand.

What happens to sessions I start in the terminal?

They appear in the same Bivy app as every other session. Leave the terminal, open the conversation from your phone, then return later without rebuilding context.

Do I need a Bivy Cloud account?

Only if you use the hosted control plane. You can self-host the control plane and relay and run Bivy without a Bivy Cloud account.

Can Bivy Cloud read my code or prompts?

No plaintext prompts, transcripts, diffs, repo files, or model keys on the hosted control plane. Remote traffic is end-to-end encrypted between your machine and devices. Account, routing, billing, and run-status metadata stay on the control plane.

Are all agents equally supported?

No. Pi, Claude Code, Codex, and OpenCode are Supported. Other adapters may be Beta or Experimental. See the agent matrix.

Is Bivy open source?

Yes. Bivy Core — the node, CLI, web app, relay, and control plane — is free and open source under AGPL-3.0-only.

Can I run more than one machine?

Yes — as many machines as you want on Self-hosted and Cloud. Pick the machine when you start work, set an automation default, or override per event.

Does Bivy sandbox every agent?

No. Some agents have their own sandbox, some expose tools Bivy can gate, and some largely run as you. Bivy does not ship an OS-level jail.

When should I use a hosted coding agent instead?

Use a hosted coding agent if you want the vendor to own the box, model bill, and most of the environment. Use Bivy when the work should run on your machines and stay something you can open and steer.

Try it

Start a session. Leave your desk. Keep control.

Run the coding agents you already use on your own machines. Start them manually or automatically, then continue from terminal, browser, or phone.

$ curl -fsSL https://bivy.sh/install.sh | bash

Read the quickstart → View on GitHub → See where Bivy fits →