Feature idea: VS Code extension powered by Odysseus

[IvanNikitin2]

Hey everyone,

I've tried using Odysseus as my local AI workspace and noticed it already has a solid REST API with Bearer token auth. This got me thinking about creating an official VS Code extension that brings Odysseus directly into the editor.
A VScode extension connects to your running Odysseus instance (http://127.0.0.1:7860 by default) and exposes AI assistance right where you code.

Some core ideas:

• Inline chat - select a block of code, press a shortcut, ask a question, get an answer in a panel beside your editor, basically how Github Copilot works, but with self-hosted or API from Odysseus.
• Sidebar chat panel - allows to change models, ask questions about the repo, make changes, check diffs, and run commands. Streaming API replies.

The good news is the backend already supports streaming responses via /api/chat_stream, so the hard part is already there. The extension would just be a thin VScode layer on top.

Would anyone else find this useful, and would the maintainers be open to it?

[CoolJohn-lab]

Holy shit yes. Maintainers on this repo can't stop you doing it, just do it...

There are some VS Code plugins like Cline and Continue that will hook up to the models you can serve on Oddy already. Just get some coder model going via the Cookbook, and point Cline at your local model endpoint, bing bang bong, you're coding agentically in VS code.

Cline is crappy though, there's room for much, much better VS Code plugins. Stuff that hooks into the Oddy ecosystem and can access your memories and whatnot. Or you could do a Cursor and fork VS Code into an entirely new thing, and hook it up to a bunch of bespoke coding agents that you create in Oddy, the sky is the limit!

[Michiel-VandeVelde]

can't stop? i certainly wont, This is a cool idea,
I’d open an issue for it so people can find it, discuss the idea, and jump in if they want to help.
And if you start prototyping, a draft PR or separate repo would be awesome :)

But feel free to approach it however you want!

[Cinabutts]

came here in hopes this idea's been posted, inspiration that brought me here: https://github.com/yoy123/lmstudio-copilot-provider

[CommanderTurtle]

Yeah, I'm all for it! latest vscode changelogs showed over the last couple months allowing a custom provider. Already, as it stands, a model can be pointed to if Odysseus serves the endpoint.

An example is here (shown with insiders, on windows):

AppData > Roaming > Code - Insiders > User > chatLanguageModels.json

I put an example file up here if anyone wants an exactness for the syntax for a vllm toolcaller:

• [source code] - for the json

Future implementation would be very neat for native odysseus as the workspace manager & tool validator. Like others were saying, it would be a very incredible project

[NathanJargon]

Bumped. Very useful to a lot of devs~

[TuanKietTran]

I'm thinking the same with Zed too.