Prompt generation — a rough idea to a good prompt
Describe what you want in whatever words come out. LLMLinq rewrites it into a clear, specific prompt you can paste straight into your CLI, and shows alternatives from a curated library so you can pick rather than guess.
How to do it
- Open My work and type your rough version in the prompt box.
- Press Generate prompt.
- Compare the results — the best match is marked, and each says what produced it.
- Copy the one you want, or edit it first.
Before and after
"make my python faster" becomes: "Profile my Python code to find the performance bottlenecks, then optimize the slowest parts. Keep behavior identical. For each change, explain what it does, the expected speedup, and any trade-offs."
Which CLI writes it
Whichever you have. LLMLinq tries the tools it found in turn and uses the first that answers; pick one in the Send with row to put it first. There is no limit and no cost to us, because the work runs under your own CLI login on your own machine. This is deliberately different from sending a prompt, where the tool you pick is the only one that runs — here any tool that can write you a good prompt will do.
When one of them is broken
It moves to the next and tells you what happened. These tools fail in ordinary ways — not signed in, wrong plan, provider outage — and a silent fallback would look like your choice being ignored. A CLI that just failed is skipped for a few minutes so every generation does not wait on it, and is tried again after that.
What happens to a prompt written here
Nothing is shared unless you ask for it. Tick Submit to resource gallery beside Generate prompt and the rewritten prompt becomes a candidate — never what you typed, never your history, never your answers. A candidate is a long way from a published template: it is stripped of names, contact details, addresses, dates, payment details and credentials by fixed rules; rejected outright if it carries key material or any long token those rules do not recognise; rejected if the text still reads as being about a particular person or client; rewritten generic and polished; re-checked; then reviewed again by a separate model asked only whether anything personal survived. Every check fails closed, including one that cannot run. Most candidates are discarded, and that is the design working.
If you set your own API key
It wins, every time. Configuring a key is a statement of which model you want and you are paying for the tokens, so it is tried first and its result is the one you are handed. If the key is rejected or out of credit, you are told in the provider's own words and a local CLI covers for it — never a silent swap.
When no CLI is installed
If you have Ollama running, an open model on your machine writes it instead — weaker than a real CLI, but it needs no account. Otherwise you get matches from the resource gallery: prompts other people have already refined, found by similarity. LLMLinq installs nothing on your behalf; it uses what is already there and tells you what to install if there is nothing.