What it is. And what it isn't.
What it is
Tyfold drives the terminal you already have, under the credentials you already pay for, and it runs as many of them side by side as you want. Every pane is a real terminal, so any command works in one. It is useful before we support a tool, and better afterwards. Claude Code, Codex CLI and opencode have that today. More are coming.
What it isn't
Not agent orchestration. Nothing here dispatches sub-agents at a goal and reports back. An agent can ask for new sessions, but only you can allow them. Tyfold won't do the work. It widens how much of it you can drive at once. It doesn't host a model or hold your provider keys either. Your own CLI signs in, exactly as it does now.
The way we code changed
The tools mostly haven't. An IDE arranged around the file you are typing in made sense when typing was the job. The work arrives through a terminal now, and in most tools the terminal is still a panel at the bottom of the window. Tyfold starts from the other end.
The job is directing now
When the model writes the code, your day stops being typing and turns into directing, reviewing and deciding, across several pieces of work at the same time. You already do that from the terminal. An editor is built around the file you are typing in. Tyfold is built around the sessions doing the work, and it keeps them where you are rather than asking you to come to them.
Every session's state in one column
Working, waiting on you, or finished, without cycling through terminals.
The rail lists every session with its state: working, waiting on you, or finished. One look tells you which one needs a decision. You do not cycle through terminals to find the one that stopped.
Queue prompts while a session is busy
They are held, never sent on their own, and they are still there tomorrow.
Ideas arrive while a session is still working on the last one. Add them to a queue, reorder or edit them, then send them one at a time when you are ready. They are held, never sent on their own. The queue belongs to the project folder, so it is still there tomorrow.
Run what an agent just told you to
Into a shell, the agent's prompt or an interpreter, unsent.
Hover a command an agent printed to run it in the paired shell or at the agent's own prompt. A Python, JavaScript, TypeScript, R or SQL block in Claude Code's reply opens in the project's interpreter or database. Each can wait unsent while you edit it.
Click a file path to open it at that line
Only paths that exist become links, so ordinary prose never does.
A session mentions models.py:18‑20. One click opens the file at that span, highlighted. Every path is checked against your files before it becomes a link, so ordinary prose is never turned into one. Web links open in your browser. Right-click one instead and you can reveal it in your file manager, open it with another app, copy its path, or start a session in its folder.
Read and edit the file the agent is changing
A real editor, and it tells you when the file changed underneath you.
Open a file from a path in the output, a search hit or the tree, and read it in colour beside the session editing it. It is a real editor rather than a viewer, with find and replace on your own shortcuts. When something rewrites that file underneath you it says so rather than showing you the version from ten minutes ago, and it will not throw away an edit you have not saved.
Search the project from the same window
Search with a regular expression, or list every place a function is actually called.
Type part of a name to open a file. Search the text of every file, with a regular expression when you need one, and read the hits as they arrive rather than after the walk finishes. In Rust, TypeScript and JavaScript you can ask a narrower question: every place a function is actually called, matched against the parsed code, so its own definition and the times it is merely mentioned stay out of the list.
Review and commit without leaving the window
Stage a hunk or a single line, then commit.
The source control panel opens on the session's own repository. Read the diff of what the agent wrote, stage a whole file, a single hunk or one line, write the message, commit. Branches, stashes, tags, fetch, push and pull are there too, and the history can read another branch without checking it out.
Keep track of what your sessions cost
Your five-hour and weekly windows, and where a week actually went.
Your five-hour and weekly Claude windows sit in the status bar with the time each one resets, once Claude has reported them, and a session stopped by a limit says which window stopped it. Codex CLI's windows show there too. Beside them Tyfold keeps a running total of what it has watched you spend, broken down by project, by day and by session, so you can see where a week went. The figure is Claude's own estimate rather than a bill.
Almost nothing to set up
Pick a folder and the session is running. There is no project file to write.
Install, open, pick a folder, and the session is running. No project file to write, no config to read, no defaults to fix before you can start. The only thing you need to know already is the terminal.
One project, many sessions
Parallel sessions are usually spread across separate projects. Run several on the same repo instead. Each session gets its own git worktree, which means its own branch and its own working copy, so no session overwrites another's uncommitted changes. It is the same terminal you would have used for a single session.
Own branch & tree
Each session gets its own worktree: a separate branch and a separate checkout. Two sessions can edit the same file at the same time without conflicting.
No stashing, no collisions
Refactor in one, debug in another, review a branch in a third. All three run at the same time, and you never run git stash to move between them.
One-step spin-up
Tyfold suggests a free branch name, and you can change it before it is created. When a worktree has no session running in it, Tyfold tells you and removes it once you confirm.
Agents know they are in Tyfold
When a local session starts, Tyfold tells Claude Code, Codex CLI
or opencode that it is running in a Tyfold pane. An agent splitting work
across this repo can then ask for a worktree and a session for each
branch, in one group, through tyfold ctl or the tyfold mcp tools. Each request names the pane
that asked and waits for you to allow it. It cannot type into a pane,
and a prompt it queues waits for you to send it.
If Pro comes, pay less for it for life
Tyfold is free, with everything unlocked: no limits, no account, no card. It will stay free. A paid plan, Pro, may come later, and some of what is unlocked today may move into it. We are still figuring this out. Sign up before Pro launches and your Pro price is discounted for life if the time comes. We'll also email you when the macOS and Windows builds are out.
One more step after this: we'll send a confirmation link, and you're not on the list until you click it. We'll only use your email to tell you about major releases and to send your discount. Unsubscribe anytime. Details in the privacy policy.
Tyfold is early software, worked on every day. We build Tyfold in Tyfold, which is how most rough edges get found, and you will still meet some. Tell us in the discussions when you do: we read everything that arrives there, and it shapes what gets built next.
Download & install
Prerequisite: bring your own CLI and provider account, Claude Code, Codex or opencode. Tyfold drives them, it doesn't host a model.
Linux today. macOS and Windows next. Get an email when they land.
Get more done in the same hour.
Tyfold stays free. A paid plan, Pro, may come later, and anyone who signs up before it launches pays less for it, for life.
Lock your founder discount