Model & effort routing
Suggests a model, surface and effort across 15 AI platforms, checks its recorded status, and flags when a lighter option may be enough.
Prompt writing · 15 AI platforms
Right model. Right prompt. One result.
A control panel for prompting. Paste a rough draft. It suggests a model, surface and effort level, tells you when a model's status can't be confirmed, asks a few questions, then rebuilds the draft into a structured prompt. It suggests; you decide.
◎ Opens in ChatGPT.
- model
- claude-sonnet-5
- surface
- chat
- effort
- medium
- status
- listed ✓
01 · the problem
Too many models. Too many platforms to weigh.
You bring the idea. The maze is everything around it: a dozen models on rival platforms, several surfaces, and a range of effort levels, with no map for which to choose. So a rushed prompt meets an oversized model, and you pay to redo the output.
- Which model fits this task
- Usage burned on the wrong tier
- Re‑prompting the same idea
02 · the pipeline
Sketch in. Final precise prompt.
Four stages, one message, one thread, grounded in the idea behind the tool: “take whatever prompt it’s given and make it clearer, more specific and easier for the chosen model to follow.”
Analyse
Reads the goal, the domain and what is missing. It asks only what it needs to craft this prompt, and lists any assumption it has to make.
Route
Recommends a model, surface and effort across 15 AI platforms, checks the model's recorded status, and gives the reason. It suggests, you decide.
Rebuild
Writes a structured prompt (XML tags or Markdown, to suit the platform) with role, context, task, constraints and a verification line, keeping your goal and facts.
Thread
Every follow up refines the same prompt instead of restarting, so a whole multi step task lives in one conversation.
03 · see how it works
A common task, before and after.
One simple, everyday request. Step through it with the tabs, or your arrow keys.
Rough draft
Example rough drafts people paste, such as: help me summarise this long textbook uploaded; write a cold outreach email to a new client; make a 7 day study plan for my exam.
- No goal, audience or format stated
- No model, surface or effort chosen
- No idea what the output should actually be
What the Optimizer does
- Analyses: a faithful summary of a long text, for study, not a rewrite.
- Routes: a mid tier model is plenty. It checks the model's recorded status. It suggests, you decide.
- Frames your file: the textbook is something you attach in your chosen tool, never uploaded here.
- Rebuilds: adds role, context, constraints, output format and a verification line. Intent intact.
Precise prompt · two ways
Same draft, two results. Pick a thorough pass or a right‑sized quick pass.
<prompt model="claude-sonnet-5" surface="chat" effort="medium">
<role>Study assistant that summarises academic material faithfully.</role>
<context>
The reader will attach a long textbook or chapter in their chosen tool.
Audience: a student revising. Goal: understand, not skim.
</context>
<task>
Summarise the attached text into a 10 line overview, the key terms
with one line definitions, and 5 questions that test understanding.
</task>
<constraints>
Use only the attached text. Flag gaps. Invent nothing.
Plain language. Stay faithful to the source.
</constraints>
<output_format>Overview, then key terms, then questions.</output_format>
<verification>Note any section you could not find in the text.</verification>
</prompt><prompt model="claude-haiku-4-5" surface="chat" effort="low">
<role>Concise study assistant.</role>
<context>The reader will attach a long textbook chapter. Audience: a busy student.</context>
<task>Return a one page summary and 8 flashcards as question and answer pairs.</task>
<constraints>Use only the attached text. No invented facts. Short, clear sentences.</constraints>
<output_format>Summary first, then flashcards.</output_format>
<verification>List any topic with too little detail to summarise.</verification>
</prompt>tags model · surface · effort values guardrails. Model values are examples. The Optimizer tells you when a model's status can't be confirmed, and never forces a choice.
Why it matters
Same idea, two outcomes. A guessing model, or one that knows the goal, the source, the limits and the output, at the effort the job needs, not more.
04 · features
Three features doing the heavy lifting.
Structured prompts
Structured role, context, task, constraints, output format, plus a built in verification step.
One evolving thread
Follow ups diff and extend the same prompt, so a whole multi step task lives in one conversation.
Tips for best results
- Paste the real draft, not a description of it.
- Name the AI tool you plan to use, if you have one.
- Say the goal, the audience and the exact output you want.
- Keep refining in the same thread.
05 · changelog
What shipped, and what is next.
A living log.
Shipped
- Now
Support for 15 AI platforms, matched to the tool you name.
- Now
Model status lookup before a recommendation, with a clear note when a status can't be confirmed.
- Now
Right-sizing that can flag a lighter model when it may be enough.
- Now
Structured prompts with a built-in verification step.
- Now
Single thread follow ups, so one prompt serves a whole task.
Upcoming
- Planned
More supported models over time.
- Planned
Clearer examples on this site.
- Planned
Your idea here: suggest one through the feedback form.
06 · trust
Independent by design.
Not affiliated
Independent of OpenAI, Anthropic and every other AI provider.
Runs inside ChatGPT
It takes no actions outside the chat. A small helper server answers model-status and prompting-note lookups and stores nothing. Your conversation is handled by OpenAI under its policies.
Works from your draft
It builds from what you paste, not your saved memory, library or files.
Suggests, never forces
You keep control of the model and the wording, every time.
See exactly how it handles your data on the privacy & data page.
07 · faq
Straight answers.
Does it answer the task or write the prompt?
It writes the prompt. The model you choose produces the answer. The Optimizer crafts, improves and threads the instruction.
Which AI tools does it support?
15 AI platforms, including chat assistants, coding agents, and design and writing tools. Name the tool you use and it matches the prompt to it; if your tool isn't supported, it tells you.
How does it save usage?
If your model is heavier than the task needs, it may suggest a lighter option with the reason, then follows your choice.
Will it change my meaning?
It aims not to. It improves structure, specificity and format and keeps your goal and facts. Check the result before you use it.
Open the console. Paste a draft.
Rough idea in, structured prompt out, in one thread.
Open Prompt & Model OptimizerA plugin for ChatGPT · 15 AI platforms