thinkhard.SI
Sign in Add API key

Get better answers from AI.

Free tools for working with ChatGPT, Claude and Gemini. Fix a weak prompt or start from one of 519 templates, then make the draft sound like you, check its facts and keep private details out of the chat.

Check a prompt Browse templates

Free, with no account to make. Your prompts stay in your browser. New to this? Start here.

Your prompt 5 A start Score out of 10, updated as you type

    Graded on this page as you type. Nothing is sent anywhere.

    What do you need to do?

    Each tool does one job. Pick yours, or start from what you want to do. Most of them work without an API key.

    For everyday AI No key needed

    1. No. 1 Write a good prompt from scratchPrompt BuilderAnswer a few plain questions about the task and get a prompt that scores 10 out of 10, laid out for ChatGPT, Claude, Gemini or any other model. Free, no key needed.
    2. No. 2 Fix a prompt that gives weak or vague answersPrompt GraderPaste it and get a score out of 10, with what is missing, as you type. Free. With an API key it also rewrites the prompt for you and tests the rewrite.
    3. No. 3 Stop explaining yourself in every chatCustom InstructionsTell your assistant once who you are and how you like answers. Get standing instructions for ChatGPT, Claude, Gemini and Copilot, and where each one goes. Free.
    4. No. 4 Make an AI draft sound like you wrote itRobot-Speak CheckerPaste an email or a post and see every phrase that sounds machine-made, fix the easy ones in one click, and stop them next time. Free, no key needed.
    5. No. 5 Paste an AI answer into Word, Docs or SlackCopy CleanerTurn the stars, hashes and long dashes into real headings, bold and bullets, ready for Word, Google Docs, email, Slack or Teams. Free, in your browser.
    6. No. 6 Check an AI answer before you use itFact Check ListGet a checklist of every number, quote, source and link in the answer, riskiest first, with a search link for each. Free, no key needed.
    7. No. 7 Ask AI for help without sharing private detailsSafe PasteSwap names, emails and account numbers for placeholders before you paste, then put them back in the answer. Free, all in your browser.
    Seven more tools, for building AI into a productModel Compare, Prompt Tester, Chatbot Safety Test, Token Counter, AI Cost Calculator, Version Compare and My Prompts

    For building with AI Some need an API key

    1. No. 8 Check whether a long text will fitToken CounterPaste a document or open a file to see its tokens, words and pages, whether each model takes it in one go, and split it into parts if not. Free, no key needed.
    2. No. 9 Find out what an AI feature will costAI Cost CalculatorPick what you are building and how often it runs, and see the monthly bill on every major model. Free, no key needed.
    3. No. 10 Pick the right AI model for a jobModel CompareSend one prompt to up to six models and compare the answers, the speed and the price side by side.
    4. No. 11 Test a prompt before your team relies on itPrompt TesterRun it on a set of real examples, on one model or several, and see exactly what each change fixed or broke.
    5. No. 12 Make sure your chatbot keeps its rulesChatbot Safety TestCheck a chatbot’s instructions for weak spots, then try harmless attacks on it to see whether it can be talked into leaking or misbehaving.
    6. No. 13 See exactly what changed between two versionsVersion CompareCompare two versions word by word, and see what the edit does to length, cost and the score. No key needed.
    7. No. 14 Keep your best prompts in one placeMy PromptsSave prompts from any tool with their full history, roll back, and reuse them anywhere on the site. Stays in your browser.
    Start from a prompt that already works519 free templates for support, sales, hiring, finance, meetings, training and more. Fill in the blanks and copy. Learn to write better promptsThe free prompt course: twelve short lessons, each with an exercise scored as you type.

    Free AI prompts for your job

    519 prompts, grouped by the work they do. Each page says when to use them, what to fill in and how to check the answer. Each one works in ChatGPT, Claude, Gemini and Copilot.

    In season now: report card comments for teachers and performance reviews for managers. New to read: what a one-line prompt scores, SI vs AI, AI words to avoid, how to fact-check an AI answer, how to anonymize text before you paste it into AI and MCP servers for people who don't code.

    Used in every kind of team

    Finance teams pull assumptions out of filings. Recruiters write fair interview scorecards. Clinics draft patient letters. Each guide shows the workflow, the templates and what it costs.

    All templates All use cases Everything to read

    What a better prompt looks like

    On the left, the kind of prompt most people type. On the right, the same request after the Grader rewrote it: who it is for, what to use, what to avoid, and blanks for the facts only you know. Same task, far more useful answers.

    1Needs workBefore, 1 out of 10 (an example)
    Write a launch email for our new invoicing feature. Make it good and not too long.
    Missing Who it is for, background facts, length, format, what to avoid, what good looks like, an example, what to do if unsure, layout.
    9StrongAfter, 9 out of 10 (the same example)
    <context>
We are launching [FEATURE NAME]. What it does: [WHAT IT DOES].
The email goes to existing customers who give it about ten seconds.
</context>

<task>
Write the launch email. Its one aim is a click on Try it, so lead with the customer's problem.
</task>

<constraints>
- Body of 90 to 130 words, because most readers will not scroll.
- Use only the facts above. Where a detail is missing, leave a note in brackets.
</constraints>
    Covered Nine of the ten checks. Still missing: a sample email to match.

    Check your own prompt The score is free and instant. The rewrite uses your own API key.

    Nothing you type reaches us

    Most AI apps pass your words through their own server first. thinkhard.si only sends you the page. When you run a prompt, your browser talks to the AI company directly, with your own key. The one exception is Pro’s included AI, which you choose: it passes your prompt through our server on the way and keeps none of it.

    A typical AI app

    You Their serverreads your prompt and can keep a copy The AI

    thinkhard.si

    thinkhard.sisends the page, then steps aside Youkeys, prompts and results stay on your device The AI companyOpenAI, Anthropic, Google… straight there, with your own key
    In a typical AI app, your prompt passes through the app’s own server, which can read it and keep a copy, before it reaches the AI. With thinkhard.si, the page is sent once, and then your browser talks to the AI company directly with your own key, so your prompts never pass through thinkhard.si.

    Kept in your browser

    • Your API keys, sent only to the provider they belong to. Turn off “Remember” to keep a key for this tab only.
    • Prompts, test suites, runs and grades.
    • Drafts and settings. Export everything as one file, or delete it, from Settings.

    What reaches thinkhard.si

    • The request for each page, as with any website.
    • Anonymous daily totals of page views and tool runs. What is counted.
    • Whatever you type into the waitlist, contact or repository unlock form.

    No account, no cookies, no ads and no third-party scripts. Free keys from Google AI Studio, OpenRouter and Groq make every tool free to use.

    Free stays free. Team is next.

    What we are building next is for teams: your best prompts, tested and approved once, then one click away in everyone’s assistant. It is not built yet and has no price yet; what teams tell us decides both.

    • A shared library of your team’s prompts, with an owner and an approved version for each.
    • Approved prompts one click away in everyone’s ChatGPT, Claude or Copilot.
    • Every prompt checked and tested before it is approved.
    • Your team’s rules for private data, applied before anything is pasted into AI.
    See the plans

    Tell us what your team needs

    Four quick questions: how big the team is, what kind of team, which prompts you would share first, and what it would be worth. One email when it opens, nothing else.

    Building AI into a product? Tell us which builder tool you need.

    Fair questions

    Do I need to be technical to use this?

    No. If you can write an email, you can use the templates, the Grader’s score and the cost calculator. Underlined words open a plain explanation, and the Start here page points you to the right tool in a minute.

    What is an API key, and do I need one?

    It is a code from an AI company, such as Google, Anthropic or OpenAI, that lets these tools send your prompt to their model, billed to your own account, usually a fraction of a cent per run. You only need one to run prompts on real models. Google AI Studio, OpenRouter and Groq give free ones. The full explanation.

    Is it really free?

    Yes. The tools run in your browser, so serving them costs very little. Model calls are billed by your provider on your own key. Google AI Studio, OpenRouter and Groq all give free keys with free models, so you can use every tool without paying anyone.

    Where do my API keys go?

    They are stored in your browser and sent only to the provider they belong to, directly and over HTTPS. thinkhard.si has no server that receives them. Set a spending limit or a small prepaid balance on any key you use in a browser.

    Which models can I use?

    Claude, GPT, Gemini, Grok, DeepSeek, Mistral and open models through OpenRouter or Groq, plus any OpenAI-compatible endpoint, including local ones such as Ollama or LM Studio.

    Do I need to write code?

    No. Every tool works by pasting text. Developers can export suites and results as JSON or CSV.

    What does the score out of 10 mean?

    A count, not an opinion. A prompt meets ten checks: what to make, who it is for, background facts, length, format, what to avoid, what good looks like, an example, what to do if unsure, and a layout the AI reads well. Each one it covers adds a point. Each score has a thinking level, from Guess at 1, where the AI has to guess nearly everything, to Superintelligent at 10, where it has nothing left to guess.

    How will this make money?

    Through paid plans for what needs a server. Team, being shaped now, shares a team’s tested prompts inside the AI tools it already uses; a later plan for builders adds scheduled test runs, prompt CI and managed credits. The free tools stay free. See the plans.

    Do you train on my prompts?

    No. They never reach us. What happens to your prompts at the model provider is set by that provider’s terms. Paid API traffic is generally not used for training, but some free tiers and free models may log or use what you send, so keep sensitive data off free keys and check your provider’s policy.