Cloud agents you can watch. On the AI you already pay for.

Hand Claude Code or Codex a task. It runs on a real cloud machine you can watch, steer, and stop.Your Claude or ChatGPT subscription covers the model. No second AI bill.

Run your First AgentSee a real PR review

Loved by engineers from

Interactive demo of the Aether app, with three selectable stories. Start an agent: the new-task composer types a prompt, a devbox boots in about five seconds with a live terminal, file tree, and a shareable preview link on port 3000, and the agent fixes a broken mobile order summary, captures before-and-after screenshots and a session video into the thread, asks one question, and opens pull request #1851 with a preview link and passing checks. The review loop: aether review requests changes on a pull request, the agent folds both fixes in, and the re-review approves. PR review: the reviewer surface at /pr-reviews streams three inline findings onto a pull request and posts a changes-requested verdict. Buttons inside the demo switch tasks and pages and answer the question; the pause control freezes all motion.

app.runaether.devOpen Aether →
All tasks1 waiting on you · 3 recent
fable · 4 steps, touches 6 files12m ago
#1852+121 -9fable · aether review reading the diff10s
fix: double-charge on checkout retry#4+48 -2sonnet · Merged2h ago

Aether gives your coding agent a cloud machine you can watch, runs it on the plan you already pay for, and puts the proof on the pull request. It starts where you already work: Slack, Linear, GitHub, CLI, MCP, and Mobile.

Fix the order summary on mobile checkoutRunning
  1. Starting workspace
  2. Workspace ready
  3. Step 2 of 4 · Reproduce on a 390px viewport
  4. Running the test suite
Session videocheckout-flow0:52

A devbox you can watch.

Each task runs in its own cloud machine, and every command and file change streams as it happens. Approve the plan, steer the agent, or open your own terminal in the same box.

A warm slot starts a task in about five seconds. From there, every command streams live.

  • A fresh Firecracker microVM per task
  • Every shell command logged
  • SSH into any devbox

Your agent. Your subscription.

Claude Code, Codex, or OpenCode on your own subscription, with 40+ models to pick from. Aether only bills the machine.

See pricing

Reviews that run your code.

Turn on reviews and Aether checks every new PR in your connected repos, whoever wrote it: your agents, Claude Code, Codex, or your team.

  1. Traced, not skimmed

    Findings come from the running app, not just the diff.

  2. Before and after

    Screenshots of the base branch beside the reviewed commit.

  3. Video proof

    With video on, the reviewer records a real browser driving the change.

  4. A reviewer you can answer

    Reply on the line, or hand the fix to an agent.

See how reviews work

One task. A team of agents.

  1. Hand a big job to a lead agent.

  2. It splits the work across sub-agents in parallel, each in its own devbox, on its own branch.

  3. Their work comes back as one pull request for you to merge.

How campaigns work

Everything around the agent, built in

The machine, the proof, and the plumbing, on every plan.

  • Video proof on PRs

    Turn on demo video and the agent records a real browser clicking through its change.

  • Automated PR review

    Turn on reviews and an agent comments on every PR, then re-checks each push until it approves.

  • Live port previews

    Every port the agent opens gets a private URL you can click into and share.

  • Scheduled automations

    Give a saved prompt a schedule: every few hours, daily, or weekly.

  • CLI

    Run aether run and stream a cloud task into your terminal.

  • REST and WebSocket API

    Projects, tasks, and automations behind platform keys, with the live workspace streamed.

  • Skills from your repo

    Commit a SKILL.md and every agent that touches the repo follows it.

  • Reply by email

    Answer from your inbox and the run resumes where it stopped.

EasyScale fired itself from code review.

I used to pull down every PR and review it myself. Now it’s all done before I even look. That’s the real time saver.

Brannon HogueFounder, EasyScale

Pay for compute. That's it.

1 credit = 12 minutes on the default Small devbox (1 vCPU, 4GB). Credits meter active runtime only, per second.

  • Free

    $0

    Kick the tires on real tasks.

    Credits a month
    20
    Machine time
    about 4 hours
    Warm slots
    1
    Tasks at once
    3
    Start for Free

  • Starter

    $30/mo

    A steady lane of agent work.

    Credits a month
    400
    Machine time
    about 80 hours
    Warm slots
    2
    Tasks at once
    10
    Subscribe to Starter

  • Pro

    Most Popular

    $75/mo

    The everyday fleet for one engineer.

    Credits a month
    2,000
    Machine time
    about 400 hours
    Warm slots
    3
    Tasks at once
    25
    Subscribe to Pro

  • Max

    $120/mo

    For a fleet that never sits idle.

    Credits a month
    4,000
    Machine time
    about 800 hours
    Warm slots
    5
    Tasks at once
    50
    Subscribe to Max
  • Medium (2 vCPU, 8GB) = 2× credits. Large (4 vCPU, 16GB) = 4× credits.
  • Idle devboxes bill $0. Warm slots are free.
  • Need more? Credit packs on any plan, at $0.10 a credit.
  • No per-task fees. No overage surprises. Cancel end-of-cycle.

Common questions

Where do my API keys live?

Encrypted at rest with AES-256-GCM under a key derived per user, and injected into your isolated VM only for the duration of a session. You can revoke them at any time. The security page describes the full picture.

Is my code safe?

Each task gets its own Firecracker microVM that is never reused for another task or another user. The devbox suspends when idle, and the machine and its volume are destroyed when your pull request merges or closes, or after 48 hours of inactivity. The devbox is open the whole time, so you can watch every command your agent runs.

Do you train on my code?

No. And because the agent runs on your own provider account, what your model provider does with your code is governed by your account and its settings; we're not in that loop.

Can Aether review any public GitHub pull request?

Aether reviews pull requests in repositories where you install the GitHub App. Pull requests from forks run in review only mode.

How do credits work?

Credits meter active runtime only, per second. One credit is twelve minutes on the default Small devbox; Medium runs at 2× and Large at 4×. Idle and suspended devboxes bill nothing, and warm slots are a plan entitlement that never consumes credits.

Which agents and models can I use?

Claude Code, Codex, and OpenCode, each running on your own provider account, whether that is a subscription or an API key. Across the three you can reach 40+ models from Anthropic, OpenAI, Google, DeepSeek, and more, and you can set the model and reasoning effort per task.

Is there an API?

Yes. A documented REST API with platform keys covers projects, tasks, automations, and usage, and a WebSocket API streams the live workspace: terminal, files, agent events, ports, and git. The CLI is built on the same surface.

Can I cancel?

Yes: self-serve via the billing portal, effective at the end of your billing cycle. The Free plan is permanent, so you can also just step down to it.

Stop firing agents into the void.

Bring Claude Code or Codex, start tasks from GitHub, Slack, Linear, the terminal, MCP, the API, or a schedule, and watch every run in a devbox you can see. Free to start.

  • MCP
  • Mobile
  • Claude Code
  • Codex
  • GitHub
  • Slack
  • Linear
  • Terminal
  • API
  • Schedules
Run your First Agent