Start with your workout and booklet direction only. Once those are set, LiftRPG moves you into the build flow instead of leaving the whole system open at once.
A workout program is required
Better results: mention the kind of booklet you want, the voice/style, the mechanic feel, and whether you want quiet realism or overt genre. A few words still works.
Both available doors run the same checked Standard build: use an API key, or a coding agent on your own subscription.
A saved project is available on this device.
Use Your API
Connect your API and start the build
Your workout and booklet direction are locked in. Add your API settings, then LiftRPG will run the full build here in the browser.
Saved Progress
Step 1Start the bridge on your machine
This door uses the coding agent you already have installed, on your own subscription, instead of a metered API key. A small program on your machine — the bridge — hands LiftRPG's requests to that agent and hands the answers back. It never sees your login, and nothing is sent anywhere except by the agent itself.
Get the bridge
Download The BridgeOne file, liftrpg-bridge.mjs. You need Node installed and a logged-in Claude Code; if either is missing, the bridge’s first line names the exact fix.
This command is written for the LiftRPG folder’s own copy. If you downloaded the single file above, run it from wherever you saved it — node liftrpg-bridge.mjs, with the same flags.
You are ready when its first line readsfound Claude Code, logged in, ready on :8090
If it prints something else
The bridge checks itself before it starts serving, and its first line is either the ready line above or the exact fix. These are the three it can print instead:
Claude Code CLI not found (claude). Fix: npm install -g @anthropic-ai/claude-code
Claude Code is installed but not logged in. Fix: claude auth login
Port 8090 is already in use. Fix: stop the other process, or run with --port <n>.
If you move it to another port with --port, change the Base URL below to match.
Step 1Where should LiftRPG send the work?
Pick a provider from the list, or point the Base URL at any service that speaks the OpenAI API — Moonshot (Kimi), OpenRouter, Together, DeepSeek, or a model running on your own machine. Choosing a provider fills in a sensible Base URL; overwrite it whenever you are going somewhere else.
Running a model locally? Ollama and LM Studio both serve an OpenAI-compatible endpoint on your machine, and most local servers need no key at all.
Step 2Your key
Held in this tab for this session only. It goes straight from your browser to the provider you chose — LiftRPG has no server in the middle and never stores it. It is also what lets the next step ask your provider which models it offers.
Step 3Which model writes the booklet?
Refresh Models asks your provider what it currently offers. If your model is not on the list — or the provider does not publish one — choose Custom model and type the id yourself.
LiftRPG can load the current model list from supported providers so you do not have to keep updating this field by hand.
Step 4How should it build?
The default writes the book one part at a time, checking each part before it pays for the next — the most consistent result on most models. The timeout is how long LiftRPG waits before giving up on a single request; slow stages are allowed to take longer than this.
Advanced
LiftRPG knows list prices for the major models. For anything else — a new release, a reseller, your own endpoint — the run total honestly reads “—” rather than inventing a number. Enter your provider’s rates here and this run will show a real total instead.
Optional, and display only. These numbers change what the cost meter shows you; they never change what your provider charges.
Not every model can complete the full build pipeline. The models below have successfully generated a complete booklet on this device and are validated for use.
Provider, model, base URL, build method, and any pricing you enter are saved on this device for next time. Your key never is.
Live Run
Track the booklet build
Follow the live API run from staged generation to local validation. Completed API work, restored checkpoints, and local finish steps are labeled separately so you can tell what really happened in this run.
Keep This Tab OpenSaved progress is too large for durable storage, so it is being kept in this tab only. Closing or reloading this tab loses it — use Save Progress below to keep a copy on disk.
Waiting For Settings
Provider Setup
No request has been sent yet.
Run Progress
0 / 0 stages
Provider
No provider active yet.
Run Total
—
What Happens Now
The API pipeline will report each stage here as it runs.
Pricing And Tokens
—
LiftRPG starts with the local pricing table, then refreshes live pricing in the background when the provider allows browser access.
FinishedYour book is ready
Open it in the renderer to read and print, or download the JSON to keep.
Completefinished with a live API call in this run
Restoredreused from checkpoint instead of calling the model again
Localassembled or validated on this device after generation
Recent Run Notes
API build is ready. Add your provider settings, then start the booklet build.
Checkpoint
Guided Build
Booklet Build Status
Finish one step at a time. Copy the prompt, paste the AI reply back, then move on.
1
Foundation
2
Booklet Plan
3
Identity
4
Weeks
5
Extras
6
Ending
Current Step
Foundation
Two moves only: copy the prompt into your AI, then paste the AI reply back here.
Where You Are
Step 1 of 6
Steps Checked
0 / 6 ready
What Happens Now
Foundation
This step sets the tone, mechanics, and core structure for the rest of the booklet.
Which version this door
builds: this is the earlier generation of LiftRPG. It works start
to finish and gives you a complete, printable booklet — but the newer game
design, and the build steps that go with it, live on the API build modes. A rebuilt
guided door is planned; until it lands, this path stays as it is rather than
half-changing under you.
Before you start: the API build modes run a
game design step ahead of this one, and every step after it is written to
serve those rules. This door has no such step. It is stated in full at the Identity
step, and each step’s card (click any step above) says what this door does not
do for that step.
Step 1
Copy To Your AI
Copy the step prompt and paste it into ChatGPT, Claude, or another LLM so it can generate the booklet foundation JSON.
Show Step Prompt
Step 2
Paste It Back Here
Once the LLM has produced the JSON, paste it back here and LiftRPG will pull out the foundation data automatically.
LiftRPG could not read that response yet.
This step locks the week plan, extras, and ending logic before writing starts.
Step 1
Copy To Your AI
Copy the step prompt and paste it into ChatGPT, Claude, or another LLM so it can generate the booklet plan JSON.
Show Step Prompt
Step 2
Paste It Back Here
Once the LLM has produced the JSON, paste it back here and LiftRPG will validate the booklet plan before you move on.
LiftRPG could not read that response yet.
This step sets the title, cover, rules, and theme before the content pages begin.
What this door does not do:
the API build modes run a game design step before anything else — the
model writes the game’s rules (how you win, what you do, what the currency is,
how the password is earned) and everything after is written to serve them, with the
play spine checked against those rules both ways. This guided path skips that step,
so the game’s shape is decided here alongside everything else and nothing checks
the two against each other. It also means the weeks, found documents and endings on this
path are written without those rules in front of them: on an API build mode every
one of those stages is handed the game’s currency, its core verbs and its win
condition as a given, plus the week’s own row of the tension budget — this
door has none of that to hand them. Use an API build mode if you want the rules-first pass.
And one more: the API build
modes run an economy pacing step after the identity step, where the model says
how often the player touches each part of the economy — every week, once,
inside a window that closes, or arriving later on purpose. Each week is then checked
against that promise as it is written, so a gauge the rules feed every week and
the pages print in one week of six is refused rather than delivered. This door has no
such step and no such check, so a surface the rules assume from session one can simply
be missing from a week and nothing will notice.
Step 1
Copy To Your AI
Copy the step prompt and paste it into ChatGPT, Claude, or another LLM so it can generate the booklet identity JSON.
Show Step Prompt
Step 2
Paste It Back Here
Once the LLM has produced the JSON, paste it back here and LiftRPG will check the booklet setup before weeks begin.
LiftRPG could not read that response yet.
Focus on one approved week group at a time so fixes stay local and easy to understand.
Focus on one extras batch at a time so the document voices stay specific without turning into a full-book rewrite.
This is the last AI step. After this, LiftRPG assembles and validates the booklet on this device.
Step 1
Copy To Your AI
Copy the step prompt and paste it into ChatGPT, Claude, or another LLM so it can generate the ending JSON.
Show Step Prompt
Step 2
Paste It Back Here
Once the LLM has produced the JSON, paste it back here and LiftRPG will check it before final assembly.