Building & editing programs
Describe a program in plain language and your assistant builds it live in Protocol - as a reusable library template by default, or assigned to a client when you say so. Bulk edits across weeks work the same way.
This is where the agent earns its place. Building a multi-phase program in the dashboard is dozens of clicks; describing it to your assistant is one sentence. The assistant builds it directly and live in Protocol - there’s no draft queue and no Apply step.
The loop is just two beats - describe → it builds it, live - so you skip the assembly entirely. By default it builds a reusable library template (not attached to anyone), so nothing reaches a client until you ask it to assign.
Build from a prompt
Describe the shape you want - split, length, level, emphasis - and let it fill in the detail:
👉 Open it in Protocol to scan every session, or just tell me a client to assign it to.
It pulls real movements from your exercise library, follows your Vault conventions, and leaves the program live next to your other templates - a normal template you can open, edit, or assign.
It’s already in your Vault
There’s nothing to apply - the template exists the moment the assistant finishes. Open it in Protocol and it’s a normal program you can scan, tweak, or assign:
Want changes? Just say so in chat and it edits the template in place - “make week 4 a deload,” “swap leg press for hack squat,” “add a third upper day.” Each edit lands live; there’s nothing to re-apply.
Template, or straight onto a client
The assistant builds in one of two modes, and it picks based on how you ask:
- A library template (the default). A reusable plan that sits in your Vault, attached to nobody: “build a 4-week PPL.” Best when you want to reuse it, eyeball it first, or assign it later.
- Assigned to a client directly. Name a client and it creates the plan already live on that client’s profile - “build a 4-week PPL and assign it to Dejan.” No separate assign step; it’s on them immediately.
It can also copy an existing template onto a client - “put the Intermediate Upper/Lower block on Sarah.” All of this needs a read + write key (assigning a plan is a write, not an outward send).
Copy, close out, translate
- Duplicate into your library. “Make me a copy of the Upper/Lower block to rework.” The copy is a new private template with its own workouts and meal plans, attached to nobody, so editing it never touches the original. Workouts and nutrition templates can be duplicated the same way.
- Close out a finished block. When a new mesocycle replaces the old one, ask the assistant to expire the old plan. It moves to Expired and ends today, which is different from Paused (on hold, coming back). See Plan status.
- Dates follow the length. Moving a plan’s start date, or changing its length in weeks, moves its end date with it (start plus the number of weeks), the same way the plan settings do.
- Translate a whole program. “Translate the Shred block into Serbian.” This runs in the background, rewrites the program, its weeks and every workout and meal plan it uses in the new language, and uses your AI credits. It will not start if you are over your AI usage limit. The assistant can tell you when it has finished. See Translating a program.
Deleting asks first, and edits can be undone
When an edit would remove existing weeks from a program, or rows from a workout or meal plan, the assistant is stopped and shown exactly what would go. It should ask you before it goes ahead; a removal only happens once it repeats the change with your confirmation.
Every change the assistant makes to a program, workout or nutrition template is also saved in that item’s version history, labelled Edited by assistant, next to the state it replaced. If an edit goes wrong, open the version history in the dashboard and restore the earlier version.
Edit in bulk
Bulk changes are where chat truly beats clicking - find-and-replace across a whole structure in one line:
The same works for nutrition templates, single workout templates, and multi-phase programs.
Where chat wins - and where the dashboard still wins
Lean on the agent for the generative and bulk work; keep the dashboard for precise and visual work.
| Reach for the agent | Stay in the dashboard |
|---|---|
| Build a program/template from a description | Nudge one set from 8 to 10 reps - just click it |
| Swap a movement across every week at once | Drag to reorder a day by feel |
| Spin up three variations to compare | See a week laid out as a grid |
Changes are live - there’s no Apply step. A read+write key lets the agent build, edit, and assign directly, so everything it does is real in Protocol the moment you ask. If you’d rather review before it reaches a client, just have it build a library template first, eyeball it in the dashboard, then tell it to assign (or assign in the dashboard yourself). That template-first habit is your “look before it’s live” - not an approval queue. See Safety, scopes & privacy.
Next: put the agent to work on your roster - Analysing & summarising →