FAQ & help

Select “Start for free” or “Prompt builder” in the main menu. The original builder, its settings and batch mode are still available there.

A prompt is the task you give an AI: a question, an instruction or text with requirements. For example: “Explain conditional probability to a Grade 11 student with an example.”

1. Describe your goal in the main text field.
2. Choose the task type and language of the AI response.
3. Select a suitable structure and optional document outputs.
4. Add useful details and click “Create prompt”.
5. Review the result, edit it if needed and copy it into your AI chat. Your chosen AI then completes the task.

No. “Improve a prompt” only requires the original task. Context, audience, clarification answers, extra files and quality criteria are optional. Select only what matters. “Improve an answer” also requires the previous answer and your specific feedback.

This version uses fixed rules and text templates. It preserves your text, organises your details and adds selected requirements. It does not understand the topic like a language model, conduct research itself or produce the substantive AI answer.

“Finished prompt” shows the copyable task. “What was added?” lists the rules actually applied, such as language or table format. “Original” shows your original task; batches show the template and selected topic. Manual result edits are not automatically re-evaluated in the additions list.

Choose “Improve an answer”. Enter the task, existing answer and specific feedback. Use “What should be preserved?” for a structure or passage that already works. The follow-up prompt asks for targeted changes and a complete revised version, while still flagging conflicts or demonstrable errors.

The counter and generator count Unicode code points. Simple emoji count as one character; combined emoji may contain several characters. Line breaks count as one character. Long text is not truncated: shorten it before creating your prompt.

Open “Improve an answer” in the footer. The twelve templates and the correction builder have their own page. Reach the existing prompt builder and batch mode through “Start for free” or “Prompt builder”.

Choose the problem and click “Use in the generator”. The template fills the correction instructions. Add your original task and the previous AI answer; optionally name the parts to preserve. Review the instructions and click “Create follow-up prompt”. Copy it into your AI, which performs the revision. You can also select “Describe your own problem”.

Only the correction instructions are replaced. If that field already contains text, you must confirm the replacement. Your original task, previous answer, content to preserve and settings stay on this page. Regenerate the follow-up prompt after changes. Changing pages or website language, or reloading, discards your input; it is not saved persistently.

Below the correction instructions, you can add rules for preserving facts, missing information, using your material, deliverables, checking criteria or dividing tasks. A click appends the wording to the correction instructions, where you can edit or delete it. “Only my material” also enables the corresponding option under “More settings”. To undo it, remove the wording and uncheck the option. Nothing is inserted if it would exceed 1500 characters.

The twelve cards are original editorial wording aids. They have not yet been comparatively tested for answer quality. Each card states required inputs, success criteria and limits. A working generator is not evidence of a better AI answer. Rankings and model comparisons are not included yet.

No. It creates instructions for your chosen AI: where web access is available, open and read sources, then check whether they support each claim. A reachable link does not prove a claim. If the AI cannot check the sources, it should say so explicitly.

General task: tasks without a specialist template.
Learning & teaching: explanations, exercises and learning material.
Writing & editing: articles, emails and other text.
Research & comparison: finding, interpreting and comparing information.
Coding: identifying errors and describing changes.
Review & analysis: assessing material against criteria.
These choices add relevant instructions; they do not start independent research or coding.

“Create directly” creates a structured brief immediately. “Prepare with three guiding questions” shows three prepared questions for your task type and includes your answers. Your chosen AI is then asked to pose at most three essential follow-up questions if crucial details are still missing. Leave irrelevant questions blank.

Context is the information the AI needs: source text, learning level, figures, quantities, constraints or evidence. Audience describes who will use the result, such as “parents without specialist knowledge” or “advanced high school maths students”. Do not include confidential information.

DE/EN in the header changes the website language. “Language of the AI response” controls the added prompt instructions and requests that response language. It does not translate your original input. Explicitly requested translations into other languages are still allowed.

Hover over an option on a computer to see its explanation. You can also click “?” or reach it with the Tab key. On a phone, tap it. Click again or press Escape to close it. Change the actual selection using its checkbox or radio button.

Detail sets the desired depth: concise, moderate or detailed. Structure sets the presentation: prose, bullets, a table, numbered steps or JSON. “Fit the task” lets the AI choose. A table, for example, can be either short or detailed.

Three switches appear by the result after generation. They change the matching form options and instantly create a new version in your browser. For batches they apply to all topics. None of these actions sends your text to AGIbeamer.

No. The review brief is a second prompt, not a completed review. It includes your generated task and criteria for the task type. Copy it and replace [PASTE ANSWER HERE] with the actual AI answer, then ask your AI to process it. Unverifiable points should be explicitly flagged.

Choose “Multiple prompts”. Write, for example, “Create five learning questions about [TOPIC].” Enter Fractions, Percentages and Probability on separate lines. One click creates three separate prompts, each with a review brief. [TOPIC] and [THEMA] are replaced only in the template. Context and options apply unchanged to all topics. A batch supports 1–20 topics of up to 200 characters each; filled templates may be up to 6000 characters.

Select one result to copy or share it. “Copy all prompts” copies the batch without review briefs. “Download batch as TXT” saves all prompts and review briefs in a text file. It does not produce AI answers, Word or PDF files. Local result edits survive switching between topics and are included in copies and downloads. Regeneration replaces them; these local edits are not saved on the server.

No. The rules load with the page. Generation, instant additions and batches then run locally in your browser. No connection to a workstation or language model is needed. The site does not assess your topic semantically.

DOCX is the usual Microsoft Word document format. Choose it for a document you want to edit later. It can also be imported into Google Docs. AGIbeamer asks your chosen AI for a DOCX file; it does not create a Google Docs link in your account.

Choose PDF for a fixed page layout to read, print or share. Recipients should see the same page arrangement. Select DOCX as well if you expect substantial later editing.

LaTeX is a text-based system for typesetting documents and mathematical equations. For example, \frac{1}{2} describes a fraction. This option is useful for maths materials. The prompt requests complete LaTeX source (.tex) as an additional output; a LaTeX environment is needed to compile it into a typeset document.

Yes. The checkboxes allow multiple outputs. Your AI can be asked for an editable Word document, a PDF and LaTeX source. Equations should be typeset in the PDF and, where possible, editable Word equations in DOCX rather than raw commands. Whether the AI can actually deliver them depends on its available tools.

AGIbeamer creates a prompt: the brief for your AI. Selecting PDF adds that requirement to the brief. Your chosen AI can create the file if it has suitable file tools. Otherwise, it is asked to explain the limitation and provide the full content with export instructions. It must not invent download links.

JSON is a structured text format such as {"subject":"maths","count":8}. Software can process it. For an explanation or worksheet, choose prose, a table or steps instead. JSON specifies the data structure; DOCX/PDF specify additional document outputs.

Careful research should include opening and reading sources before citing them. This does not happen automatically in every AI response: a model can mention a link from prior knowledge or invent an incorrect address. Selecting “Request sources” explicitly asks for opening and checking sources. AGIbeamer itself does not visit links entered in the form.

Link check: is the page reachable at this address?
Content check: does the page actually support the associated claim for the relevant period and context?
Example: a university link works. That does not prove the claim “This degree programme is free”. The page must actually support that statement for the programme and period in question.

The prompt asks the AI to disclose missing web access or inaccessible pages. Such sources should be labelled unverified; the AI should not claim to have read them. You can supply a relevant excerpt or use an AI with research access. Still check important claims and evidence yourself.

Yes. It may be outdated, refer to another country or population, or contain unreliable information. It may only support part of a claim. A useful source review therefore considers the content, date, scope, publisher and availability of a primary source.

“Include an example” requests an illustration. “Explain clearly” requests essential steps, calculations or reasons. “Flag uncertainty” asks for distinctions between facts, assumptions and open questions. These are instructions to your chosen AI, not guarantees that it will fulfil them correctly.

No. Extra requirements help only when they fit the task. Use “Copy original” for a simple question that is already clear. For important work, compare both versions using the same model and assess correctness, task completion, format and editing required.

They pass the prompt currently visible in the result field to the selected service, including your local edits. Telegram also includes a link to AGIbeamer. Your reference number and a separate log file are not shared. The prompt itself may of course contain information you entered in your task.

No. The button opens the messenger’s sharing flow with prepared text. You choose the recipient and confirm sending there; you can cancel. An installed app or usable web version may be needed. The selected service receives the transferred data when you click the share button, so do not share confidential content.

The site does not shorten your text. For very long messages it avoids link-based sharing because browsers and messengers have different limits. Use “Copy prompt” and paste the full text into your messenger instead. Split it into messages there if needed.

The visible result still belongs to your previous input. Create it again to include your new choices. You can directly edit the result field and immediately copy or share those edits. All these edits stay in your browser.

Only in this page’s browser. The builder sends neither your input nor results to the operator or an AI provider. The existing builder does not persist them in cookies, Local Storage or IndexedDB. Exception: the additional “Improve step by step” page lets you optionally save its history in this browser; see the new FAQ section. Closing or reloading the page can lose your text; save results yourself. Browser extensions, browser features and your own clipboard are outside this website’s control.Contact enquiries are sent to our server and forwarded by email only when you submit the contact form. Only the transfer button puts a packet in the same browser tab’s Session Storage. It is valid for ten minutes and is removed on import, discard or the next access after expiry. The browser may keep it until then or the end of the tab session. It is not sent to AGIbeamer or put in the URL. The offline preview uses memory only.

Only the UTC time of the page request and a source category: for example Google, Bing, a known network site, internal, other, or direct/unknown. Classification uses the browser-supplied referrer. Its full address, search terms, IP address, location, browser identifier and visitor ID are not stored. A category may be missing or inaccurate; it does not prove the visitor’s origin.

STRATO processes technical connection data to serve the website. There is also the minimal access log, external links, the contact form and email and messenger sharing. These flows still need to be described. No content storage does not mean the whole website operates without data processing.

Seven calendar days by default. The configured period is stated in Privacy. Old files are removed on page requests or by a maintenance job. The entries count page requests rather than unique people; bots may be included.

Prompt files saved before the update are no longer appended to. The operator must review old records and backups and remove data no longer needed. The existing expiry period still applies to remaining legacy files. New prompts do not receive record IDs.

Copying places the selected text on your clipboard. TXT downloads create a file on your device. Clicking WhatsApp or Telegram passes text to that service; you choose the recipient and confirm sending there. Sharing is optional and subject to that provider’s terms.

Check required fields and length limits. Open “More details” when needed. After a website update, reload the page completely. If clipboard access fails, the site selects the text for manual copying.

Open “Contact” in the menu or footer, choose a topic and write your message. The form forwards your details by email to prompt@agibeamer.com. It does not copy any text from the prompt builder. In the offline and setup preview you can try the form without sending anything.

The form code saves no message copies or contact database. Your enquiry is handled by email. A private technical security file contains only a signing key, short-lived random identifiers and submission timestamps for rate limits; no names, email addresses, messages or visitor IP addresses. The privacy page provides details.

They ask for details relevant to a study plan, explanation, exercises or email and turn them into a prompt. They do not call an AI model. Missing information remains an open point; your AI should ask about crucial gaps.

Click “Continue with this prompt” below the result. The exact visible version, including your manual edits, is transferred. For a batch, this means the selected task. Review the preview and confirm the import on the destination page.

Only after you confirm replacing it. Export any history you need as TXT first. If local history saving is enabled, the confirmed import also replaces the saved version. Cancelling preserves the existing history.

Only the transfer button puts a packet in the same browser tab’s Session Storage. It is valid for ten minutes and is removed on import, discard or the next access after expiry. The browser may keep it until then or the end of the tab session. It is not sent to AGIbeamer or put in the URL. The offline preview uses memory only.

Explicitly selected rules: word limit, simple list count, exact strings, headings on separate lines, certain placeholder patterns, and pure JSON with specified top-level keys. Differently formatted content can be missed. Facts, calculations and curriculum alignment are not verified.

A string may be written or formatted differently. “Exact string not found” does not automatically mean “wrong”. A matched amount may also be used in the wrong context. Decide yourself whether correction is needed.

In the improvement history, turn flags into feedback, review it, then create a new prompt version. If it exceeds 1,000 characters, choose fewer checks. On the separate check page, copy a correction prompt together with the answer.

Choose a course and material at PunkteMachen. Enter the topic, level, excerpt and difficulty here yourself. The learning assistant creates a prompt for your own AI. It does not automatically read accounts, PDFs, quiz progress or curricula. LearnForReal remains a separate entry to quizzes.

Not for AGIbeamer. PunkteMachen, LearnForReal and your chosen AI may have their own access requirements. Merely opening a link does not automatically transfer your content between services.

Only an explicitly saved improvement history and a still-valid transfer packet in the same tab can remain available. Assistant inputs and standalone requirements checks are not saved persistently. Copy needed results or export the history as TXT.

Create a prompt, test it in your AI and paste its answer here. Choose what needs improvement. AGIbeamer creates the next prompt version while preserving your other requirements.

Yes. AGIbeamer is free without sign-up and does not call a language model. Your chosen AI has its own terms, costs and usage limits.

No. You judge its content and quality. AGIbeamer counts words and characters and identifies some predefined conflicts between requirements. It does not perform semantic AI analysis or complete fact-checking.

It is associated with the correct prompt version. You can select a problem passage and compare answers. Only the latest answer is included as material in the next prompt; older answers are not repeatedly appended.

You can stay in the same chat or start a new one. The prompt includes your goal, requirements and, from version two, the answer to revise. Files and images from earlier chats are not transferred; attach them again in your AI if needed.

Preferences in other categories are retained. A new choice in the same category replaces that category’s earlier choice and additional details. Old versions remain in the history. Free text is not semantically reconciled; review the prompt yourself for complex requirements.

Open “View my versions” and choose “Continue from here” on an earlier version. Later versions remain in the history. You can then try a different improvement.

By default it exists only in the open tab. Enabling “Remember the history on this device” saves this service’s task, prompts and answers in this browser’s Local Storage. There is no synchronisation between devices, browsers, languages or domains. Private browsing and clearing site data may remove it.

“Start over / delete history” clears this language’s history from the page and local storage. Disabling local saving removes the saved copy but retains the open history. “Save history as TXT” downloads a local text file. Delete downloaded files and clipboard contents yourself if needed.

Up to 20 prompt versions per history. A task may contain up to 6,000 characters and an AI answer up to 20,000. Overlong input is not silently truncated. Save a TXT copy before starting over.

No. The prompt builder, batches and repair templates remain available. “Improve step by step” is an additional independent page for several revision rounds.

You paste text here. Files, images and model selection stay with your chosen AI service. AGIbeamer does not access your AI accounts or automatically send prompts to them.