gallanon
Prompt anonymization: fictional details and restoration of AI replies
You are starting a new letter, claim or other document. Type or paste your prompt with real details into gallanon, review the selections and mark anything missed. You receive a prompt with fictional details and a key. Send only that prompt to AI. Then save the AI response in a supported file, load it with your key in gallanon, review the matches and download the response with the real details restored.
How it works, step by step
- Select the third mode and enter your instruction, circumstances and details needed for the draft. Text stays in the current tab's memory.
- Run detection. Review names, addresses and numbers, and manually select anything missed, including unusual circumstances.
- Generate the prompt after reviewing it. gallanon randomly assigns replacements and records the mapping in a key. Read the result before copying it to an LLM.
- Save the key, preferably encrypted. Send only the prepared prompt to AI and ask it to preserve fictional names and numbers.
- Save the AI response as TXT, MD, RTF, DOCX or ODT. You can load it and the key in the restoration workflow even after restarting your computer.
- Review proposed restorations, confirm name variants and map missed passages. Download the document containing real details.
The difference from the first mode is what the model sees. Document mode uses explicit labels such as [PSN-1]. Prompt mode uses fictional names and values, so the instruction resembles an ordinary case description. That does not guarantee a better AI response.
Replacements are generated randomly on your device. The key records mappings, observed forms and selected inferred name variants. Confirm uncertain restorations; these variants do not cover every grammatical form.
The response can be an entirely new document. Its filename and number of occurrences need not match the prompt. Keep the correct key: a fictional name alone cannot recover the real one.
Example use
You want a payment demand but have no draft yet. Write an instruction with the parties' details. AI receives the fictional version. Load its new letter and your key in gallanon to restore the actual parties.
What to check
- Fictional details do not prevent identification through the remaining story. Remove or generalize unnecessary circumstances and review the whole prompt.
- Changing dates, amounts or facts can affect the model's reasoning. Do not use fictional dates to calculate actual deadlines or replacement numbers as valid official identifiers.
- AI may shorten a name, use only a first name or invent someone else. gallanon recognizes stored variants, not every paraphrase. Potential unmatched details are an aid, not a complete list of AI mistakes.
- Restoring names does not verify the draft's substance or legal reasoning. Read the finished document before using it.
Pseudonymized data remains personal data. A redacted document may also identify someone through its remaining context.