Run coding agents
on your own server.

Hand tasks to Claude Code, Codex and OpenCode. Close your laptop. They keep working.

Free · macOS and Windows · Any server you can SSH into

● My server · Connected
Retry failed webhooks with backoff, and add a test.

Start a task tonight. Wake up to a pull request.

Always on

Close the laptop. Agents keep working.

Connect your own server in a few clicks. Agents run there overnight.

✓ Ubuntu 24.04 · x86_64
✓ Reaches GitHub and your model providers
! 1 GB memory: one agent at a time
✓ No ports opened: private connection over SSH
Approve and installabout 2 minutes

The research

Your agents' day is 24 hours long.

Agents can now work for hours on their own. The biggest gain isn't typing faster next to one. It's the work that gets done while you're away.

16 more hours of agent work, every day A laptop runs agents while you're at it. Your server runs them around the clock: 24 hours instead of 8, three times the work time.
  • On your server
  • On your laptop, while you work
Agents now work for hours on their own Length of task the best agents finish half the time, measured in how long it takes a professional. It doubles about every seven months.
Working side by side with AI: a modest gain Controlled studies of developers coding with AI help. Faster on some tasks, slower on others. Delegating whole tasks is where the hours come from.

Sources: METR, Measuring AI Ability to Complete Long Software Tasks and METR's time-horizon results to April 2026 (50% success; the 80% horizon is several times shorter, so review still matters) · Peng et al. 2023 · Cui et al. 2025 · Paradis et al. 2024 · Becker et al. 2025 (METR).

Parallel

A workspace for every agent.

Each agent gets its own copy of your repo. Run as many as you like.

Retry failed webhooksbranch thread/retry-webhooksWorking
Upgrade payment SDKbranch thread/payment-sdkDone · 6 files
Fix flaky cart testbranch thread/cart-testNeeds you
Dark mode for settingsbranch thread/dark-modeWorking

Your agents

Bring your own subscriptions.

Use the agents you already pay for, with your skills and tools.

AgentClaude Code
ModelSonnet · 1M context
EffortHigh
frontend-designgithubcontext7+ your skills
Claude CodeCodexOpenCode

Teams

From one agent to a team.

A lead agent splits the work and reviews the results. You approve what matters.

Leadplans and reviews
API changesdone
Frontendworking
Ship it?waiting for you

Loops

Agents on a schedule.

Nightly reviews, weekly updates, issue triage. They only ping you when it matters.

Nightly code reviewevery day · 02:00All good
Dependency checkMondays · 08:002 updates need you
Issue triagewhen an issue opensLabelled 14
Next run: switch the reviewer to a lighter model · −38% cost

Mobile

Conduct from your phone.

Scan a code to follow your agents and reply from anywhere.

Retry failed webhooksWorking
Tests pass. Should I also retry on HTTP 429?
AllowReject
Yes, and cap it at 5 attempts.
Reply to the agent…

Review

Review, then ship.

See every step and every change. Merge only what you like.

src/webhooks.rs +12 −3
let resp = client.post(url)- .send().await?;+ .send_with_retry(5).await?; record(&resp).await;
Copy to my checkoutOpen pull request

Safe by default.

Isolated

Agents can't touch your machine or your keys.

Your rules

Choose what agents can do and which sites they reach.

Budgets

Set a limit. Agents stop when they reach it.

Full history

Every action is recorded and can be exported.

Free.

No account, no subscription. You only pay for the models you use.

Get started.

Free. No account.

  1. Open Threadrun. It sets itself up.
  2. Sign in. Use your Claude or ChatGPT plan, or a free model.
  3. Describe a task. Press Enter.
First launch

macOS: unzip, move Threadrun to Applications, then right-click it and choose Open (the app isn't notarized yet).

Windows: unzip and run threadrun.exe. If SmartScreen warns, choose More info → Run anyway.

Questions

Do I need an account?

No.

Which agents can I use?

Claude Code, Codex and OpenCode.

What does my server need?

Any Ubuntu server you can SSH into. Threadrun sets it up for you.

Can agents reach my machine or my keys?

No. Each agent works in its own sealed workspace.

Is the phone link safe?

Yes. It needs a PIN that only your app shows.