Trainer

What it does

The trainer is Luna’s training coach: it works from what you actually logged and what your plan says.

• Reads back your week against your plan, and says when a streak is at risk.
• Tells you what today’s session is, or that it’s a rest day.
• Gives specific advice on sets, load, tempo, form and recovery.
• Talks through changes to your plan. The change itself is made by plan drafting and only goes live when you save it.

What it reads

• Your message and the conversation so far.
• Your health journal, body records and training plans, including logged sets and synced steps.
• Your inner context, which changes what it suggests but is never quoted back.
• Long-term memory of earlier conversations.

The server decides what goes into the prompt before the model runs. The trainer is one of only two aspects allowed to see your health data.

What it produces

A conversational reply, as text or read aloud. Numbers it quotes (sets per body part, streaks, loads you logged) come from your records, not from the model. Every exercise in a plan is a real entry from the 868-movement catalogue, and LUNA never derives a weight you haven’t lifted (see Training coaching).

Human oversight

• You save the plan. Nothing it suggests becomes your plan until you save it.
• You log the sets. Loads enter the system only when you mark a set done.
• It can’t book, schedule or change anything outside LUNA.

Model

Qwen3.6-35B-A3B, by Qwen, licensed Apache-2.0. Runink domain adaptation for this agent is planned; this release uses the base model.

Form review and rebuilding a plan from photos use a separate vision model, not this one. Both are described in Training coaching.

Where it runs and data handling

On Runink-operated servers, not on your phone. No third-party AI service is called. Your training data is stored in your own encrypted partition. On the phone apps, spoken replies are read aloud by the phone’s own speech engine. See Privacy and data.

Guardrails

• Not a doctor or physio. If you report pain, numbness or a suspected injury, it tells you to stop and see a professional before any programming.
• It won’t plan training through pain or an injury. It offers to train around it once you know what it is.
• It won’t write dosing or cycle protocols for performance-enhancing drugs, and points you to a doctor who knows your bloodwork. The topic itself can still be discussed.
• It trusts your logged history and never invents a workout you didn’t report.

See Safety for the full list.

Limitations

• It doesn’t diagnose or rehab an injury. A limit you state is treated as a limit, not a diagnosis.
• When a reading is missing, it says so rather than guessing.
• Free chat replies stream as they are written, so they aren’t checked by the judging ladder the way plan proposals are.
• Its limits are written into the prompt. They are guidance, not medical safeguards.

Evaluation

No published evaluation scores yet.

Illustrative example

Invented data.

You: How did my training go this week?

Luna (trainer): Three sessions: Monday, Wednesday and Saturday. Your plan has four training days, so you’re one short. Your streak is at three days and today isn’t logged yet, so a short session keeps it alive.