Chat with your VegaDuta agents, run workflows, and (once the on-device engine
lands in the chat view) get free local inline completions - without leaving
VS Code.
Build
The chat UI is the shared webview bundle from clients/shared - build it
first, then the extension:
cd clients/shared
npm install
npm run build:webview # -> clients/shared/dist/webview/
cd ../vscode
npm install
npm run build # bundles dist/extension.js + copies the webview into media/webview/
npm run typecheck runs tsc --noEmit over src/ and ../shared/src.
npm run package produces the .vsix (requires a prior build).
Run / smoke test (F5)
Open clients/vscode in VS Code and press F5 (Extension Development
Host; the launch config runs the build first).
In the dev host: open the VegaDuta activity bar view -> Sign in
(device flow: browser opens against Keycloak, approve, the notification
closes). Settings default to production
(api.vegaduta.ai / auth.vegaduta.ai); switch to staging or a custom
deployment via vegaduta.environment.
Pick an agent and send a message - the reply must stream with spaces
between words (the SSE data: slice(5) live check).
Run a workflow from the view or VegaDuta: Run Workflow… - status polls
every 2s until COMPLETED/FAILED/CANCELLED.
Select code in an editor -> right-click -> Explain Selection - the
selection lands in the chat composer, press Send.
(After M2) In the chat view, download the small model; the status bar
flips to VegaDuta: Local and inline completions start coming from the
on-device engine.
No sign-in available? VegaDuta: Use API Key accepts a vmcp_ scoped key
(admin-issued). Chat is blocking (no streaming) in that mode.
Settings
Setting
Default
Notes
vegaduta.environment
production
staging | production | custom
vegaduta.customApiBase / customAuthBase
–
required when environment is custom
vegaduta.completions.provider
auto
auto | local | hosted (v1: not yet available) | off
vegaduta.edge.enabled
true
allow the chat webview to run the on-device model
vegaduta.ollama.baseUrl
–
optional local OpenAI-compatible server for the engine layer
