Using AI assistance — apprenti.dev Docs

AI in apprenti.dev is assistive, never authoritative. It can explain, hint, quiz, and coach a reflection. It cannot approve your work, and on most tasks it cannot hand you the finished solution.

Two different places, two different jobs

•AI tab — a chat grounded in your whole active curriculum, not just today's task. Ask about the path, another task, a competency, or how two terms fit together. You can optionally tap Ask about {title} to bring the current task into the conversation, but the tab is never locked to it.
•Task detail → AI assistance — five task-specific modes: Explain, Hint, Quiz, Coach, and Solution. What's available here is controlled by that task's AI policy; anything not allowed shows as disabled rather than being hidden, so you know it was a deliberate choice, not a bug. See the five AI assistance modes for what each one actually does.

More → AI is where you configure a provider — it is not the assistant itself.

The guided stance

When you're chatting without pinning a task, the AI tab stays in a guided stance: it will explain and hint, but it will not produce a full assignment solution, even if you ask directly. A task's own solution-generation setting only comes into play once you've explicitly scoped the conversation to that task — and even then, most tasks keep it off, with a mentor able to override for that apprentice if appropriate.

Local or Cloud — always labeled

Every reply is tagged Local or Cloud · {provider}, matching whichever actually ran. AI can run against a local model (GGUF, downloaded from a hosted catalog or imported from this device — never fetched from GitHub/GitLab or committed to a curriculum repository), or OpenAI, Mistral, Anthropic, or OpenRouter.

You choose the routing policy in More → AI: Local only, Prefer local, Manual, Prefer cloud, or Cloud only. If no cloud key is configured, cloud simply stays off — there is no hidden fallback.

Privacy

Nothing about apprenti.dev's AI leaves this device unless you've configured a cloud provider and you're not in Local only mode. Only the active task's context (and anything you've written into it) is included when you invoke an assistant — not your whole history, not other tasks, not other apprentices' work. API keys and tokens never touch Git, the local database, or logs.

Keeping an answer

Any assistant reply — or a selection from it — can be saved with Save as note onto the current task or any other task in the curriculum. That note becomes an ordinary Git-backed file you can edit afterward; it is not a locked transcript. The chat conversation itself stays local to this device and is not something you commit — saving a note is the deliberate act that makes an answer durable.