Synod for Visual Studio Code

Synod for Visual Studio Code puts an AI chat and agent panel inside your editor. It talks to models served from infrastructure you control, not a third-party service: on your own machine today, and to Datum Core, the Photon Datum platform, once signing in with your membership is ready.

It comes with the Developer membership. Availability: Pending. The extension is in development and is not yet released to members. Each feature below says which stage it is at.

In development

A chat panel in your editor. It sits in the side bar or in its own tab, with replies that stream as they are written, formatted text and code blocks you can copy.
Named conversations, several at once. Switch between them, rename, clone and delete them; a deleted one can be brought back. They are kept on your machine and resume where you left off. Several chats can be working or waiting for a model at the same time: each is marked, and the tab shows how many.
Your choice of model. Pick from the models that are actually available, turn the model’s reasoning on or off, and see how fast it is writing.
Your editor as context. The open file, your selection and folders you choose are available to the assistant, and you control what is attached.
Tools, with levels. The assistant can read your files, git history and project tracker, and check things such as the date and time. It can make changes only when you allow it, and each change waits for your approval.
Steer an answer while it is being written. Add to your prompt and the assistant gets it at its next step, or add and restart: the answer in progress is stopped, the steps already done are kept, and it asks again with your addition.
Auto-tune. It offers only the tools a message needs and turns reasoning off for a simple question, so a small model keeps room to answer. A line above the reply says what it decided.
A record of every change. Each file the assistant changes is listed with the file before and after, and a View diff opens the comparison.
Commands in the message box. Summarise a long conversation so it fits the model’s window, see which tools and models are on offer, check how the last reply went, flag a reply that went wrong, and keep your computer awake while a reply runs.
Hooks. Add your own starting context to every conversation with a short script.
Your own tool servers. Add servers that give the assistant controlled access to other systems (MCP), and switch groups of their tools on or off.
English, German, Spanish and Bulgarian. The interface follows your editor’s language. The German, Spanish and Bulgarian text is a first draft that has not yet been reviewed by a native speaker.

Roadmap

Sign in with your Photon Datum membership and use Datum Core, our member platform, from the editor.
The distributed collaborative AI compute network and your personal and shared knowledge index, from the editor.

You stay in charge

Tools are off, read-only, or ask first. The default is read-only. When the assistant wants to change something, you see exactly what it will do (for files, a dry run first) and nothing happens until you approve. Deleting and pushing are not offered by default.
Autonomy is something you set. By default it asks each time. You can let it keep reading on its own, or let it make changes without asking only inside folders and projects you name. Every change made without asking is shown and written to a log on your machine.
What a tool returns is data, not orders. Text that comes back from a file or a page is never passed back to the assistant as an instruction by default, and an answer that looks like a command is never run.
Your conversations stay with you. They are stored on your machine. The error log never holds your prompts or the replies, masks keys, is private to you, and can be switched off.