All docs

Docs › Studio

Use Studio together with VS Code

Studio for thinking, asking, running and keeping results; your editor for reading and hand-editing code. How to get the best of both on the same folder.

Two tools, two jobs

Studio on your computer can work on an existing code base: a space names a folder, and its Code tab is a chat with your own AI in that folder, beside what changed and the project's Build and Test buttons. It does not replace your editor. The best results come from giving each the job it is good at.

JobWhereWhy
Deciding what to build, and whyStudio: Discuss, Research, NotesThe reasoning stays with the project, not in a chat that scrolls away.
Asking the AI for a changeStudio: the Code chatIt starts with the space's goal, its notes and its research results.
Reading and reviewing code line by lineVS CodeA real editor: jump to a definition, search, compare side by side.
Small edits by hand, debuggingVS CodeFaster by hand than by asking.
Build, test, runStudio: the project's command buttonsOne click, and a failure goes straight to the AI with its output.
What runs on your machinesStudio: ServicesEach service checked, its log, start and stop, in one place.
Results, numbers, reportsStudio: Research and DocumentsThey outlive the conversation that produced them.

Set it up

  1. In Studio, open the space for the project. In its view switch, open ⋯ and choose Code base….
  2. Name the folder: on this computer, or on a machine your ssh reaches. Studio looks at it and proposes the Build and Test commands; change them if they are not right.
  3. Open the same folder in VS Code. Both now work on the same files and the same git repository; nothing is copied into Studio.

Studio's Code chat runs your own AI coding agent in that folder. Your code goes to that AI service under your account with it, as it does from your editor, and not to PriorStates.

A working day with both

  1. Start in Studio. Open the space and read where things stand: the open research topics and the last results. Talk the next change through in Discuss before any code is touched.
  2. Ask for the change in Studio's Code chat. It already knows the space's goal and notes, so you explain less. For anything that matters, ask it to look and say what it would do first.
  3. Review in VS Code. Read the change there, fix small things by hand, and step through anything that looks wrong.
  4. Back in Studio: build, test, commit. If a command fails, Ask the AI to fix it sends its output to the chat.
  5. Keep what you learned. Add the result to the research topic, and put any decision in the space's Notes. Next week's work starts from there.

Moving between the two

  • Open in VS Code. In the Code tab, ↗ Open in VS Code opens the folder. In What changed, the ↗ beside a file opens it at its first changed line. A file and line the AI names in its answer, such as src/quote.py:42, is a link that opens there. A folder on another machine opens through VS Code's Remote SSH.
  • Studio sees what you did in the editor. When you come back to Studio's window, What changed is up to date. If files changed outside the chat since the AI's last turn, your next request tells it which ones, so it reads them again before relying on what it remembers.
  • One conversation in both places. A conversation you started in the editor's AI extension can go on in Studio: in the Code tab, ⋯ → Continue a conversation from VS Code… lists the folder's recent conversations. Studio's own Code chat is an ordinary conversation in that folder too, so the extension's history can pick it up on that machine.

These need Studio for computers 0.5.37 or later. Opening in VS Code needs VS Code on the computer Studio runs on. Continuing a conversation works with Claude Code, in the folder where the conversation was started.

Three habits that matter most

  • One AI conversation for one piece of work. If you ask Studio's chat and the editor's AI extension about the same change, each misses what the other did. Pick one per task: Studio's when the task needs the project's context, the extension's when you are deep in one file.
  • Keep decisions in the space's Notes, not in chats. Long chats are shortened as they grow; Notes are read by every later AI session in the space. One line such as "prices are whole cents, never fractions" saves saying it again.
  • Let the AI propose before it changes. Ask it to look and say what it would do first. In Studio, Change the code on a research topic starts that way.

What stays in your hands

  • The AI leaves its changes in the working folder for you to review. It is told not to commit, push, deploy or touch a production machine unless you asked for exactly that.
  • Revert, Commit… and Push are Studio's own buttons, and Push asks first. Your editor's git tools see the same repository.
  • This is an instruction to the AI, not a lock: it runs with your own access in that folder, as it does from your editor. Review before you commit.