By Nils Ekström, CTO at Stomme AI
The phrase "your data never leaves your device" has become a marketing cliché. Vague, often untrue, and almost impossible to verify. When it comes to AI assistants, the honest answer is usually: some of it does, some of it doesn't, and the details matter.
We'd rather give you the real picture.
What stays on your Mac
Conversations. Every message between you and your agent is stored on your Mac. The conversation history lives in your local environment — not in our database, not in a cloud sync. When you search your history, your agent searches your disk.
Files you share with your agent. If your agent reads a document, analyses a spreadsheet, or writes a file, that work happens on your machine. The files don't move to our servers.
Agent state and memory. Your agent's working memory — context about your preferences, ongoing tasks, skill configurations — is stored locally. This includes structured event logs written to local JSONL files with no external network calls. Retention is configurable; default is 30 days.
Skill files. The skills that define what your agent can do sit on your Mac. If our servers went down tomorrow, your agent would keep working with everything already installed.
What Anthropic processes
Your agent is powered by Claude, Anthropic's large language model. When your agent generates a response, your message is sent to Anthropic's API and a response is returned. This is how any Claude-powered product works.
What this means:
- The content of each conversation turn is processed by Anthropic's infrastructure
- We have a Data Processing Agreement (DPA) with Anthropic covering GDPR requirements
- Anthropic's API terms prohibit using customer data to train models by default
- Neither Stomme AI nor Anthropic retains that content for training without your explicit consent
What we process during setup
When you onboard, we collect information to provision your environment: hardware specs, skill selections, agent preferences. This data is processed server-side during provisioning. After your environment is live, we don't retain the raw onboarding data beyond what's needed for your account record.
What we hold in our infrastructure
- Account and billing data. Name, email, subscription tier, payment records — stored in our Supabase-backed database. Standard for any subscription service.
- Consent records. Append-only consent logs per user — email integration, calendar access, telemetry, marketing, AI training. You can revoke any scope at any time.
- Service metadata. Operational telemetry for reliability — skill versions, provisioning status, error signals. Not conversation content.
Remote access for support
If you need help and we need to look at your environment directly, we can only do that with your explicit approval. Access is not persistent — it must be approved per session, with defined scope. Your Mac isn't streaming logs to us.
Your agent doesn't depend on us
Your agent is self-contained. Skills, configurations, and state are on your machine. If our servers became unavailable, your agent would continue to function. This is a deliberate design choice — your agent is yours, not ours.
Summary
| What | Where |
|---|---|
| Conversations | Your Mac |
| Files shared with agent | Your Mac |
| Agent state and memory | Your Mac |
| Telemetry events | Your Mac (local JSONL) |
| Skill files | Your Mac |
| Messages to Claude (per turn) | Anthropic's API |
| Account and billing | Stomme infrastructure |
| Setup data | Processed server-side, not retained |
| Remote access | Requires your per-session approval |
What we're still working toward
We're working toward publishing our full data retention policy, making consent controls accessible via the customer dashboard, and giving you a one-click data export. If you have questions, email [email protected].
Your agent runs on your Mac. Your data stays local.