visorOne VM per session.
Linux sessions run in Firecracker micro-VMs with a full Docker daemon, Compose, headless Chromium, code-server, Node, pnpm, Go, git and gh.
Cloud-first agentic development environment
A heads-up display for parallel coding agents.
Visor runs Claude Code and Codex in isolated cloud VMs, one VM per session. Each one boots with your repo, dependencies and services already running. You and your team watch, steer and review from a browser or your phone, and the work keeps going when you close the laptop.
- running
fix-auth-redirect+42−7 - simulator
ios-onboarding+118−23 - PR open
billing-webhooks+260−91 - needs you
flaky-e2e+9−4
The Conductor-alternative checklist
What people ask for, and where Visor answers it.
Taken from public threads of developers listing what they need from their next ADE.
Cloud-first. Keeps going with the laptop closed.
Every session runs in a cloud VM. Your laptop is a window onto the work, not the machine doing it.
A provisioned VM per workspace.
Each session restores its own VM from your environment's snapshot, with repo, dependencies and services already up.
The real Claude Code and Codex harnesses.
Sessions run the Claude Code CLI or Codex inside the VM, with model and reasoning effort set per agent.
Multiplayer.
Tasks belong to the account, not the creator. Send a teammate the task link and they open the same task and reply in the same thread.
See which agents are waiting on you.
Tasks that need somebody sort to the top with an amber accent and name whose answer is missing. Answer a parked question from the lock screen.
An iOS app.
Plus iPad, Android and web. Messages you send while a VM boots are queued and delivered in order.
Ports and a live preview.
Every service port in your manifest gets a preview in the task, with a selector when there's more than one.
Scheduled routines.
Playbooks run every day, weekday, week or month at a time you pick, or on events like an issue being assigned.
Real GitHub integration.
Agents push and open PRs under their own GitHub identity. A labeled issue can start a task, and a playbook can wait for an approving review.
A CLI agents can drive.
Boot VMs, run commands and start tasks from the REST API or the CLI.
Built for a team
One thread per task, and everyone on the account can reach it.
Hand off by link.
A task's URL stays valid across sessions. Anyone on the account who opens it can read the transcript and send the next message.
Questions find their owner.
An agent that needs a decision parks and asks: in a Slack thread, to a named person, or on the task. The reply resumes the session with everything it had loaded.
Credentials have an owner.
Logins live in an account keychain and are granted to agents by name. Agents type them into pages without the value ever appearing in chat.
How it works
Describe the machine once. Every session boots from it.
- 01
Define an environment.
A manifest names your repos, platform, setup commands, services and health checks.
- 02
Snapshot it.
Visor prepares the VM and saves a snapshot. Credentials are scrubbed before capture, and capture stops if that cleanup fails.
- 03
Start tasks in parallel.
Each session restores its own VM from the snapshot, with services started, on its own branch. Multiple sessions run in parallel across different repos.
- 04
Steer, review, ship.
Chat with the agent, watch the diff update live, preview the running app, then open the pull request.
What you get
Everything an agent needs, on a machine that isn't yours.
A simulator per agent.
Every macOS VM runs on Apple Silicon with Xcode and its own iOS Simulator, so parallel agents don't share one simulator or your disk. It streams live at 30fps, and you can tap and swipe.
PRs that come back checked.
Agents run the changed flow, attach permanent screenshot links to the PR test plan, and wait on CI themselves. A reviewer agent can record its findings on the pull request.
A diff that keeps up.
Each session's branch is compared with base as the agent works. You get unified or split view, a file tree, j/k to move between files, and a Create PR button once there are changes.
The app, running.
Agents take screenshots of any page your environment serves, read the rendered DOM and console errors, and click through to the state they need first.
A browser for the agent.
A real Chromium runs inside the VM and stays open across calls, so the agent can log in, click through a flow and check the result, on your services or on the open web.
Agents with their own accounts.
Agents commit under their own GitHub identity, take requests in Slack threads, and have their own email address.
Files that outlive the VM.
Attach your own S3 bucket to Linux environments, and agent output survives after a session's VM is torn down.
Cost per agent.
Model usage is reported per agent through the API or CLI, so each agent's cost sits next to what it shipped.
Visor and Conductor
Same idea, different machine.
Conductor runs Claude Code, Codex, Cursor and OpenCode in parallel, each task in its own workspace on your Mac. Visor runs Claude Code and Codex agents in a cloud VM per session, shared across your team. The Conductor column uses only what conductor.build states.
| Conductor | Visor | |
|---|---|---|
| Where agents run | On your Mac | Cloud VMs, one per session |
| Coding agents | Claude Code, Codex, Cursor, OpenCode | Claude Code, Codex |
| Isolation | A workspace, branch, files and terminal per task | A VM per session, with its own Docker daemon and services |
| Laptop closed | Runs while your Mac is awake | Keeps running in the cloud |
| Teammates | Not described on conductor.build | Account-wide tasks, shared by link |
| iOS work | Your local Xcode | macOS VMs, each with its own streamed simulator |
| Where you watch | Mac app | Web, iPhone, iPad and Android apps |
| Review | Diff, pull request, merge, archive | Live diff, then a GitHub pull request |
Pick Conductor if you want everything local, Cursor or OpenCode in the mix, or your Claude or ChatGPT plan driving it. Pick Visor if your agents need their own Docker, databases, browser or simulator, keep working when you close the laptop, and have to be shared with a team.
Questions.
What is an ADE?
An agentic development environment: where coding agents do their work and where you supervise it. In Visor that means a cloud VM per session, plus the apps you watch from.
Can my team share sessions?
Yes. Tasks are scoped to the account, so any member with the link can read the task and follow up in the same thread. A transcript can hold messages from several members, agents and Slack senders.
What happens if a VM dies mid-task?
A turn that loses its machine retries on fresh VMs, up to twice. Queued messages survive the interruption, and an idle session reboots from its snapshot on your next message.
Does my code leave my machine?
Yes. Your repo is cloned into a cloud VM. Secrets are encrypted with keys held in a separate key service, and the API returns secret metadata rather than values.
Do I need a Mac?
No. Most work runs on Linux VMs, and you drive them from a browser or a phone. macOS VMs are there for Xcode and the iOS Simulator.
Can I bring my existing setup?
Your setup becomes an environment manifest: repos, setup commands, services and health checks. On Linux, anything you can install with apt or run in Docker works in the VM.
Is Visor affiliated with Conductor, Anthropic or OpenAI?
No. Visor is independent. It runs Anthropic's Claude Code and OpenAI's Codex as coding agents.
Early access
Get your agents off your laptop.
Tell us what you run today. We'll reply with access and help turning your setup into an environment.
You're on the list.
We'll be in touch at the address you gave us.