Skip to main content
The extension’s runtime runs entirely on your machine. Outbound network requests go only to the AI provider you configured, carrying only the data listed below.

​
Sent to the model

  • Log snippets the agent has captured during the active session
  • Source files and Kconfig the agent has opened in this task
  • Tool-call results (build output, flash output, device responses)

​
Never sent

  • Files outside the active workspace
  • .env files, credentials, signing keys, board UUIDs or MAC addresses
  • Background workspace scans, idle file watchers, or telemetry from the agent runtime itself
Bring your own key: you control which model and endpoint you trust. The source is fully open and auditable.

​
Local models work

Any OpenAI-compatible endpoint can be configured, including locally-hosted models via Ollama, LM Studio, or llama.cpp’s built-in server. This is useful for privacy-sensitive projects where data cannot leave the developer’s machine. A model with strong native tool-calling is required; small local models often fall short.

​
Telemetry

Pseudonymous product analytics only (installs, activations, feature and free-tier usage, detected platform, and execution errors), keyed to a random install ID with no account, name, or email. We never collect your source code, chat content, or device logs. Opt out anytime: set telemetry.telemetryLevel to off in VS Code settings.