Skip to main content
Three places your data can be, and what each one holds.
Your machineThe runtime, your project memory, your logs, your free-tier allowance. All of it stays here.
Your model providerReceives only what the task in front of it needs: the code and log snippets it is working on.
Adsum, if you registerYour email and which features are unlocked. Nothing about your projects, ever.

You choose the provider. A local model through Ollama or LM Studio keeps everything on the machine.

​
What leaves your machine

​
Nothing runs without you

Every file write and every command waits for your approval, unless you turn auto-approval on for a category yourself. A message you type while a run is working is queued for the agent’s next step. It is never taken as an approval.

​
On your disk, in your repository

  • Project memory lives in an .adsum/ folder beside your code: the board, the transport, the goal, open defects, notes. Plain files, under your version control, never uploaded. Delete the folder and the memory is gone.
  • Compaction, when a long session is condensed, happens in the model conversation you already configured. No new destination.
  • Session export writes one file, redacted of secrets, so you choose what leaves the machine and where it goes.

​
The optional account

A free account is a key to the knowledge that needs one: cellular, satellite NB-NTN, nRF54 edge AI, and the LEW840x gateway firmware. Sign-in happens in your browser with GitHub or email. The extension never sees a password, and the token stays in your operating system’s keychain. Deleting the account removes everything keyed on you and leaves this machine’s projects, logs and allowance untouched. See your free account.

​
Telemetry

Pseudonymous product analytics only: installs, activations, feature and free-tier usage, detected platform, execution errors, keyed to a random install identifier with no account, name or email. Never your source, chat content or device logs. Opt out any time by setting telemetry.telemetryLevel to off in VS Code.

Models and providers

Which endpoint your prompts reach, and how to point Adsum at a local model.

Limitations

What the agent and the CRA readiness check do not do, stated plainly.