Talyn - Run Claude Code and Codex in parallel cloud sandboxes
by•
Talyn runs Claude Code and Codex in parallel cloud sandboxes, on the plan you already pay for. Close the laptop and come back to a diff, or to a pull request that is already green. Free with 10 box-hours a month. For GitHub, on macOS, Windows, Linux, or the browser.


p/htmldrag
p/runevr
p/producthunt
p/google-cloud-platform
p/kodlo
p/ai-context-flow
p/jot-2-0
p/beamfy
p/dex-by-exmergo
p/refs
p/reso
p/duckfightclub
p/seshka
p/nudgeforme
p/talanoa-2
p/dino-kids
p/nexupira
p/many-displays
p/murror
p/supabase
Replies
Hey everyone, Tom here. I made Talyn.
These days Claude writes most of my code, and my job had turned into babysitting it. Several Claude windows open, laptop lid up all day, and the minute I walked away a PR would fail CI and sit there red until I got back.
So Talyn moves all of that to the cloud. Every task gets its own sandbox with its own branch, agent and terminal. You can start a bunch of them, shut the laptop, and come back to diffs. (A mobile app is coming soon too.)
It runs Claude Code and Codex on your own Claude or ChatGPT plan. I didn't want to pay for tokens twice, so Talyn only charges for the machines.
The part I use most is what happens after the PR opens. Talyn fixes failing checks, conflicts and review comments, and if you queue the PR, it merges once it's green. It won't merge anything you didn't ask it to.
I also didn't want an agent holding my keys. Tokens stay outside the sandbox and get added on the way out, so there's nothing in there to leak.
A few things you should know: it's GitHub only for now and it's a beta built by one person, so expect rough edges, and please tell me when you hit one. Curious what you'd hand off to an agent first!
@tom_owers "close the laptop and come back to a green PR" is the line that got me. I build my whole SaaS with Claude Code. how do you handle secrets in the sandbox, like a DB URL the tests need?
@raghav_21 All the secrets are handled outside of the sandbox, and we use a proxy to then append secrets/variables where it matters.
@tom_owers red CI first, easy. it's the most boring part of the day and the agent doesn't get annoyed by it 😅 keeping keys outside the sandbox is a nice touch
tokens staying outside the sandbox and only added on the way out is the detail that makes me ok leaving it running overnigth
when it fixes a failing check on its own, do i see the diff before it pushes??? or does it just push and let the next CI run decide???
@thamibenjelloun Hey, with auto fix enabled, it'll push and let the next CI run. This is an opt-in feature though, and so with it disabled it won't fix it push any changes for you.
does it ever run out of github api budget? all of it rides on my user token now if i read the architecture doc right
@bbueno Nope - we track the token budget and make sure to never go over. We also use github webhooks to recieve as much data as possible so we're not polling
ah ok, that helps
where do i see it when claude stops to ask a question?
@leoheo Claude will ask a question in-line - but, if you're using a Thread, then the orchestrator will take care of these questions for you
@tom_owers ok gonna try a thread then
congrats!! what does it do when two of the parallel ones touch the same file, who wins
@igorblink It uses git to manage all the merge conflicts - so usual merging will apply here!
Hey Tom, congrats on the launch. Talyn looks really interesting. You're right, I'm spending most of my time babysitting Claude Code, and I think this would be an area that would be nice to automate. Looking forward to checking it out.
@nathan_tarbert Lovely - let me know if you have any feedback!
The babysitting part is exactly where my day goes too. One question before I try it: when you spin up parallel sandboxes, do they share any repo context, or does each agent re-read the whole repo from scratch? That re-read cost is what usually kills parallel setups for me. Also curious whether a sandbox can run an Xcode build and simulator tests, or if it's Linux-only for now.
@jcbuilds Hey, yep so the sandboxes have a shared memory that they read and write on every session. We also bake a "golden" image with your learnt setup (such as dependencies, drivers, etc) so that the sandboxes start with everything you need
github only is the one thing stopping me tbh, we're on gitlab. is that on the list or way later?
@digi_dinos Yes, absolutely - it's on the roadmap and will be coming very soon!
@tom_owers nice. gitlab.com only at first, or self-hosted too? self-hosted is usually where these integrations get stuck
@digi_dinos I can certainly get working on self hosted too
@tom_owers that'd make it an easy yes for us tbh. thanks for following up, good luck with the launch 🙌
Coming back to a pull request that already passed instead of a wall of failing checks is the part I would actually feel, since the babysitting happens while I am away from the desk anyway. Does it keep the box warm between runs so the next one does not pay for setup again?
@karimbenkeroum Yes, absolutely - so the boxes learn your set up and pre-install everything and then cache the box. So each startup gets faster and faster with your latest changes already pre loaded
Running Claude Code and Codex in parallel cloud sandboxes while automatically fixing CI failures is useful. Can Talyn enforce a maximum number of repair attempts and notify developers when a PR keeps failing instead of repeatedly consuming compute?
@mahmoud_shawkat Yep! It has a set limit of 3 attempts before it'll request your input