Sessions
A coding agent with its own sandbox.
A session is one task on one branch, with its own agent, chat, diff, files, terminal and run scripts, in a cloud sandbox. Run several at once on your Claude or ChatGPT plan, and open the pull request from the same screen.
- Its own branch and sandbox, so parallel agents never collide
- Chat, diff, files, terminal and run scripts on one screen
- One button to open the PR, fix checks, resolve conflicts and merge
- Claude Code or Codex, on the subscription you already pay for
Free: 10 box-hours a month and two sandboxes at once. Pro: 200 box-hours and up to 100 at once.
Add CSV export
sundial/web · dana/add-csv-export
I will add an export helper, put a button above the table, and cover it with a test.
Ask the agent… (@ for files, / for commands)
What a session is
A session is one task. Pick a repository, describe what you want and press Enter. Talyn names the session, makes a branch, starts a sandbox and gives the task to your agent.
The agent is the real Claude Code or Codex command-line tool. Plan mode, slash commands and skills work the way they do on your laptop.
Add CSV export
sundial/web · dana/add-csv-export
I will add an export helper, put a button above the table, and cover it with a test.
Ask the agent… (@ for files, / for commands)
Everything on one screen
The chat is in the middle. The panel on the right shows the branch, and a terminal dock sits below it.
- Changes: the diff against the base branch
- All files: browse and read any file on the branch
- Pull request: its status, checks, reviews and place in the merge queue
- Sandbox: what the sandbox is doing and whether your work is saved
- Setup, Run and Terminal: your scripts, their output, and a shell
Keep talking while it works
A message you send during a turn waits, then goes in when the turn ends. Each message you send is a checkpoint, so you can put the files back to how they were before it.
Open a second chat in the same session when you want another line of work on the same branch.
From diff to merged
One button in the header moves the session on. First it reads Create PR. The agent writes the title and description from your changes and your PR template, then Talyn commits, pushes and opens the pull request.
After that the same button offers the next step: resolve conflicts, fix checks, merge. When you are done, archive the session. Archiving saves any work you did not push.
Quick to start, free when idle
Talyn builds an image for each repository with its setup already run, plus your own setup steps and dotfiles. A new session starts from that image, not from a clean machine. Talyn also suggests setup steps it sees your sessions run again and again.
A sandbox that is idle for 10 minutes goes to sleep and costs nothing. Your next message wakes it. While it sleeps, you can still read its changes and files from Talyn's copy.
Your work is saved
After every turn, Talyn saves the work you have not pushed to its own storage. It pushes nothing to GitHub to do this. If a sandbox is lost, the next one restores the work and tells you when it was saved.
Run your app
Start your dev server with a run script. On the desktop app, Talyn forwards the sandbox's port to the same port on your machine, so you open it on localhost like any other dev server.
Local when you want it
On the desktop app, a session can also run in a git worktree on your own machine, with your own tools. You can also move a Claude Code conversation from your computer into a sandbox and carry on there.
Questions, answered.
Claude Code on your Claude Pro or Max plan, or Codex on your ChatGPT Plus or Pro plan. A sandbox session uses one agent, which you pick when you start it.
Two sandboxes at once on the free plan and up to 100 on Pro, counted across all your workspaces.
Its sandbox uses box-hours while it is awake: 10 a month on the free plan, 200 on Pro. A sleeping sandbox uses none. The model runs on your own subscription, so there is no token bill from us.
No. Saved work goes to Talyn's own storage, not to GitHub. A push happens when you open a pull request, or when the agent pushes because you asked it to.
Yes. Talyn reads setup, run and archive scripts from .talyn/settings.toml, and also from conductor.json or .conductor/settings.toml, so existing scripts work unchanged.
Talyn does this for you.
Coding agents in cloud sandboxes, from the first prompt to the merged pull request, on the Claude or ChatGPT plan you already pay for. Free with 10 box-hours a month.
Keep reading
Feature
Threads
One goal across repositories
Read moreFeature
Fix a PR with an agent
Failing tests, conflicts, review comments
Read moreFeature
Agents & Talyn Fleet
Your own Claude or ChatGPT plan
Read moreCompare
Talyn vs Conductor
Conductor runs several coding agents in parallel workspaces on your Mac. Talyn runs them in the cloud against the pull requests those agents produced. Different halves of the same day.
Read the comparisonCompare
Talyn vs Cursor cloud agents
Cursor's cloud agents run tasks in a VM and hand you a merge-ready pull request, billed at API pricing. Talyn manages the pull requests you already have and runs its agents on the subscription you already pay for.
Read the comparisonCompare
Talyn vs T3 Code
T3 Code is a free, MIT-licensed control surface for the agent CLIs already installed on your machine. Talyn runs the work on its own hardware when you are not there. The difference is not features: it is who has to be present.
Read the comparison