CodeInOven is a desktop workstation for building software with coding agents. It coordinates the agent through a lifecycle you can follow: say what you want, read the plan, approve the direction, and review the change before it lands.
The project, the conversation, and the record of what the agent did sit in one window. You do not need a second terminal for the commands or another tool for the diff.
What you need before you start
- A computer running macOS on Apple silicon, Windows, or Linux. CodeInOven is a desktop app and does not run on a phone or tablet.
- The app itself, from the download page.
- A folder for the work, when the work belongs to a repository. A chat needs no folder.
- A model to answer. Pi comes with the app, so the first message needs no other install.
A network connection is part of the picture when the model answering you runs at a provider. A model on a server you run yourself works over your own network instead.
The four first moves
Open a project or start a chat
Choose a project when the agent needs to work in a repository. Add project opens a folder you already work in, and Clone repository takes a Git URL. Each project keeps its own threads, files, and runs, and its first thread opens as soon as the project is added.
A chat needs no project. Use Chats for research, a file check, or a question that does not belong to a folder you are working in.
An empty view offers its own start list: Add project, Clone repository, Open settings, New chat, and a tour of that view. The list changes with the view you are in. Once a project is in the sidebar, hover its row to start another thread in it.
Pick the agent
Pi ships with CodeInOven, so you can start without installing anything or signing in anywhere. Settings from the rail lists every other agent the app knows. Picking one opens its accounts when this machine already has it, or the page that installs it when it does not. The composer's model control lists the models from every agent you have connected, so the choice can change as the work does. Agents covers accounts, keys, and local model servers.
Set the permission level
The composer holds one control per decision: the chat options menu, attach, the permission level, and the model with its thinking level. The permission level decides what the agent may do without asking you first. Auto Review runs a step unless it is explicitly denied, and Full Access approves every operation.
Set the level before you send. A level you change during a run applies to the next turn, so the run already in progress keeps the permission it started with.
Review the work
A run reports as it goes. The working trace lists the steps and the commands it ran, and the Changes panel shows the files it touched with the additions and deletions. Read both before you decide the work is done, and send another message in the same thread when something needs another pass.
When the agent needs a decision from you, it asks in the thread and the run waits for your answer. The question also appears on the bell, so you can leave the window while a long run works and come back when it stops for you.
The rail and the header
The rail runs down the left edge of the window and switches views: Projects, Threads, Chats, Assistant, and Browser. A gold square marks the view you are in, and each view keeps its place while you are away. The bottom of the rail holds the update status, the task manager, Utilities, and Settings.
The header names whatever is on screen. The arrows beside it walk back and forward through everywhere you have been, view changes included. The right side holds the actions that belong to the view you are in, and the bell. A finished run, a question waiting on you, and work that needs review each land on the bell, named after the thread it came from.
Next
Install the app from the install page, then follow a folder through projects. The composer page covers the four decisions in detail.