FAQ

Do I need to use the terminal?

Twice, and only for setup. One line installs Claude Code and signs you in, one line starts torta. You come back to the terminal for Ctrl-C to stop everything.

Is torta agnostic about which AI tool it uses?

Not today. Every agent torta starts is a Claude Code session on your own login. Inference is billed to your own account; no key belongs to torta. Sessions run behind a runner interface with four implementations.

What does it cost me in tokens?

Torta itself costs nothing. The sessions cost whatever your Claude Code login would cost for the same work.

Two things keep a run bounded:

• a per-ticket budget: 30 minutes of wall clock, 200000 tokens, 200 tool calls and 8 continuations, all settable in torta.json
• a concurrency cap of 3 sessions at once, plus a 60 second cooldown per agent after a session ends

An idle engine costs nothing: a session begins only for a ticket you approved.

A ticket delivered into a board pane gives up the token and tool-call budgets. For bounded spend, use --runner sdk.

Why is there no separate watcher command to run?

torta start runs the board and the engine in one process. torta watch is there for a server with no board, and for --runner stub or --runner sdk in CI and scripts.

Does anything leave my machine?

The board binds to 127.0.0.1 and refuses cross-origin requests, on the HTTP routes and on the WebSocket handshake alike. What does leave is what any Claude Code session sends: the prompts and file contents the agent reads, to Anthropic, on your own login.

Can I edit tickets by hand?

Yes. They are markdown files with frontmatter, and the board updates when you save. Two rules: the filename carries the status, and the required frontmatter fields have to be present.

If you are working in an agent’s home yourself, tell the engine:

torta watch --busy frontend

What happens when an agent gets stuck?

It stops and asks in the needs-you lane. A hard constraint parks the ticket in BLOCKED; a ticket that cannot be built as described becomes NOT_FEASIBLE; a budget that ran out parks it in BLOCKED too, naming the limit and the cost so far.

Can I run two projects at once?

Yes, on different ports:

torta start ~/projects/one
torta start ~/projects/two --port 5600

One engine per workspace, though: a second engine is refused, naming the command already running.

Does it work on Windows?

The board reports whether this machine can host a pane at all. Node 18 or newer and python3 on PATH are needed everywhere.