LinuxCommandLibrary
GitHubF-DroidGoogle Play Store

turri

Coding agent that regenerates a project from plain source files

TLDR

Start an interactive session in a project
$ turri
copy
Check that this machine can run it
$ turri doctor
copy
Write a starting model config
$ turri init-config
copy
Parse a plain file, with no model call
$ turri parse [app.plain]
copy
Show what is left to regenerate
$ turri drift [app.plain]
copy
Regenerate code from a plain file
$ turri regenerate [app.plain]
copy
Regenerate one functional spec
$ turri regenerate [app.plain] [app:add-task]
copy
Pick models for this session
$ turri --harness-model [provider/model-id] --coding-model [provider/model-id]
copy

SYNOPSIS

turri [OPTIONS] [COMMAND] [ARGS]

DESCRIPTION

turri is a terminal coding agent. The npm package is @turri-dev/cli. The command is turri. It needs Node.js 22.19 or newer, Git, and Bash, on macOS, Linux, or WSL. A desktop app lives in a different repository and is not this command.With no arguments inside a project it opens a conversation. In an empty directory it picks a stack (the one you name, otherwise a standard-library stack this machine can build), writes a module, and regenerates it. Further changes are asked for in the chat.Without a conversation, three commands work on a plain source file (a path such as app.plain):- parse checks syntax and structure and does not call a model- drift reports what is not regenerated yet and starts nothing- regenerate turns the plain source into jobs and runs them. A second argument names one functional spec.doctor checks the environment. init-config writes a starting model configuration.Flags:- --harness-model — model you talk to- --coding-model — model that writes code- --stack NAME — stack for this session- --debug — keep agent logsInside the chat, these are typed as slash commands, not as subprocesses: /login, /model, /turri-config, /halt, /lineage, /btw, /sentinel, /stack, /pulses.Three model roles exist: harness (the conversation), coding, and judge. A model is provider/model-id, with an optional :high or :low suffix for effort. One left unset falls back to the role before it: judge from coding, coding from harness.

CONFIGURATION

Project settings are config.turri.yaml. A stack can have its own file, config.turri. plus the stack name plus .yaml. /turri-config edits the project file from the chat.

CAVEATS

Asking to try a build runs release, which copies what was built into dist/ under the module name, and activate, which starts it and installs what it needs. That path does not ask for a separate approval. The project also documents a sandbox that wraps agent commands. Read that before pointing turri at a repository you cannot rebuild.

SEE ALSO

npm(1), node(1), git(1)

RESOURCES

Copied to clipboard
Kai