Talyn: Run Claude Code and Codex in parallel cloud sandboxes
Talyn is a desktop and web app where a developer describes a task and hands it to Claude Code or Codex running in a cloud sandbox. Each task gets its own branch, a sandbox started from an image of the repository in seconds, a diff view and a terminal. One button opens a pull request, and Talyn keeps that pull request passing checks until it merges. Product Hunt listed the launch on November 15, 2026, under the Open Source, Software Engineering, Developer Tools and MCP categories.
Agents run on the Claude or ChatGPT subscription a user already holds, so there is no API key to paste and no metered token bill, according to the project's own documentation. Talyn charges for sandbox time instead. The free plan includes 10 box-hours a month, Pro includes 200, and Max costs $100 a month for 1,000. One box-hour is one standard box running for one hour. A box sleeps after 10 idle minutes and uses nothing while asleep, and unused hours do not roll over. US-dollar credits cover usage beyond the included hours.
Setup has also changed recently. A pull request in the project repository replaces manual token creation with a single button in the desktop app that connects both agents on the user's computer. It writes the Talyn server into each agent's configuration and installs remote-development skills. The Product Hunt listing, the company's website and its documentation are the sources for the figures above.
Why it matters
For builders, the product sells compute rather than tokens, which puts the margin question on sandbox hours. The maintainers have published their own cost model in a pull request, estimating a likely cost of about $0.02 per standard box-hour and a cautious figure of $0.04. That same change says the Max tier cuts the per-hour price to $0.10 from $0.20 on Pro. Founders selling agent infrastructure should expect buyers to compare per-hour compute prices across vendors.
For investors and operators, the notable shift is that agent usage is billed against a user's existing subscription. That lowers the cost of trying the product, but it also ties the vendor to the terms of those subscriptions and to model providers' willingness to let third-party tools drive their agents. Teams funding or selling coding-agent tools should test how their revenue holds up if those terms change.
Key points
- 1Each task gets its own branch, cloud sandbox, diff view and terminal, and can open a pull request.
- 2Agents run on existing Claude or ChatGPT subscriptions, with no API key and no metered token bill.
- 3Free includes 10 box-hours a month, Pro includes 200, and Max costs $100 a month for 1,000.
- 4A sandbox sleeps after 10 idle minutes, and a sleeping box uses no box-hours.
- 5A repository pull request adds one-click connection of Claude Code and Codex to the desktop app.