DOCS / PRIVACY & DATA

Your host runs the AI. Memory stays local.

No DontDumb backend receives your code, prompts or answers. No DontDumb model processes your work. The learning harness uses your existing coding host’s model connection. Your learner memory and history are stored locally.

The actual data path

Selected work + relevant learner context
             ↓
Your existing host/model connection
             ↓
Learning result → local panel and memory

Local does not mean offline inference. Selected context can include captured code excerpts. Those excerpts reach your existing model provider, under that connection’s policies. OpenCode can use an API-backed provider that you already configured.

Data destinations

Data	Destination
Profile, goals, answers, drafts and history	Local files/host plugin storage. Relevant context may enter learning calls.
Selected work and captured code	Your existing model connection, not a DontDumb application backend.
Anonymous usage counts, timing and error categories	Direct to PostHog when configured. No code, prompts or answers.

Anonymous analytics

Anonymous product metadata is on by default when configured. Disable Share anonymous usage in Learn settings. Analytics is separate from model inference and local diagnostics.

Local storage

Shared learner memory is under ~/.local/share/dontdumb. Codex and Claude have host-specific companion directories; OpenCode also persists session state in its plugin storage. Exports can contain goals, answers and selected work context—keep them private.

Permissions and isolation

Review native plugin/hook prompts. The coding task stays independent. The Codex readiness fallback carries one fixed, rate-limited control instruction, not lesson or learner data. DontDumb is an additional local component with access to selected work; existing-provider reuse does not mean zero additional risk.