AI training plan
An AI marathon training plan, built from your own runs.
Ask the AI you already use for a marathon training plan and Memory.fit gives it what a template cannot: your last months of runs, your sleep and recovery, your thresholds, and the race you chose. The AI writes the whole block, with phases and a taper, into Memory.fit, and each run lands on your watch with its steps and targets.
Memory.fit makes no plan of its own. There is no 12, 16 or 20 week table here. Your AI decides the length from the weeks you have before race day and the training you have actually done, and the plan shows in the app with your AI's name on it.
Beginner or experienced, first marathon or fifth: the AI reads your history and builds from there, and you can change anything by asking.
How the block is built
Your AI starts with get_workouts, get_training_load and get_athlete to see how much running you really do, how your weeks look and which paces you hold. Then it adds your race with set_race, marks it as your A race, and writes the block with set_training_plan: phases, weekly structure and a taper toward race day.
Each session is written with plan_session and set_session_structure, so a long run with a marathon pace block becomes warm-up, work steps and cool-down with targets in your own pace zones.
Many runners search for a 12, 16 or 20 week marathon plan. The length that fits you depends on how much you run today. Ask your AI how many weeks it needs to build from your current weeks, and it can say so from your own data. If you have more time, it can add a base phase. If you have less, it can say what is realistic.
- Race date and target, saved with set_race and set_race_target
- Phases and taper, saved with set_training_plan
- Key sessions marked with set_session_key
- Sessions sent to your watch with send_session_to_devices
How it adapts week to week
A plan is a first draft. Each week, ask your AI to check in: it calls get_updates to see the runs that arrived since last time, compares them with the plan using get_session_compliance, and reads sleep and recovery with get_sleep and get_daily_metrics. If a long run was cut short or sleep was poor, it can move a session with move_session, skip one with skip_session, or rewrite the next days, and it tells you why.
Every change appears in the Memory.fit inbox and can be undone. You can also ask your AI to run the check on a schedule, for example every Sunday evening.
Because the block lives in Memory.fit, you can open the Plan tab and see every week, with the long runs and key sessions marked. Your AI can write notes on each session, for example why a run is easy.
Fuel for long runs and race day
For long runs your AI can write a fuel plan with set_session_fuel_plan: carbohydrate per hour, fluid and sodium, in your numbers. Sports nutrition guidelines give carbohydrate ranges per hour that rise with the length of the effort (ACSM position stand), and your AI can adjust them to what your gut handled in training.
For race day, Memory.fit has fuel plan pages for marathons, such as the Berlin, London, Chicago and Valencia marathons. Open one, or ask your AI to use set_race_plan for your own race.
Which watches get the runs
Memory.fit sends structured runs to Garmin, COROS, Suunto and Wahoo, and to intervals.icu. Each step arrives with its target in your own zones. A session your AI changes is updated on the watch, and your recordings come back so the AI can compare each step with its target. See the send workouts page for the details of each watch.
In your AI
The AI you already use, with your data in Memory.fit.
- ChatGPTChatGPT plans your week, fuels your long sessions and writes it all back to Memory.fit.Set up ChatGPT
- ClaudeClaude writes your plans back to Memory.fit, and asks you before it replaces or removes something you have.Set up Claude
Grok BotGrok Bot plans and reviews from your real training, and can re-plan a week after a bad night's sleep.Set up Grok Bot
MuseMuse plans sessions from your real training and writes them into Memory.fit's Plan.Set up Muse
DotsOpenAI says a dot's background research uses your connected apps read only, so plans are written when you ask in the chat.Set up Dots- GeminiType @ in a Gemini chat and pick Memory.fit to ask about your training. Gemini asks you to allow each session it plans, and planning a whole week is not reliable yet.Set up Gemini
- GrokGrok reads your training, sleep and fuel, and can plan around them.Set up Grok
- Perplexity ComputerPerplexity Computer reads your training in Memory.fit. Whether it writes a plan back isn't checked yet.Set up Perplexity Computer
- Claude CodeClaude Code reads your training and writes the sessions it plans back to Memory.fit, asking before each change unless you allowed it.Set up Claude Code
CueYour Cue agents plan your week and write your fuel plans from your real training, sleep and recovery.Set up Cue
What to send
Copy one, send it to your AI.
“Read my last 12 weeks of runs in Memory.fit and tell me how many weeks I realistically have to build to my marathon on the date I give you.”
Tools:
get_workoutsget_training_load“Add the Valencia Marathon to Memory.fit as my A race with a target of 3:30, then write a training block with phases and a taper.”
Tools:
set_raceset_race_targetset_training_plan“Plan next week in Memory.fit around five runs, make Sunday's long run the key session, and send each run to my watch.”
Tools:
plan_sessionset_session_structuresend_session_to_devices“Check in on my week in Memory.fit: what did I run, how did I sleep, and should anything move?”
Tools:
get_updatesget_session_complianceget_sleep“Write a fuel plan in Memory.fit for my 32 km long run on Sunday, in grams of carbohydrate per hour.”
Tools:
set_session_fuel_planget_session_fueling
Questions
In short.
Yes, and it writes a better one when it can read your real runs. With Memory.fit connected, ChatGPT, Claude, Grok Bot, Muse, Dots or another MCP AI reads your training and writes the plan back.
No. Memory.fit has no template plans. Your AI builds the block from the weeks you have and the training you have done.
Your AI sets the starting point from your own history. If you are new to running, tell it so and it can build a gentler block.
Structured runs go to Garmin, COROS, Suunto and Wahoo, and to intervals.icu. Each step arrives with its target.
Ask your AI to check in. It reads what happened and can move or change the next sessions, and you can undo any change in the app.
Sources · Checked November 10, 2026
Your next chapter starts with everything you've learned.
On Android now. iPhone is coming: join the waitlist and we'll email you the day Memory.fit is on the App Store.
No spam. One email when it's ready.