Private upload and redaction

Upload and donate an AI conversation

Upload a ChatGPT, Claude, Gemini, ShareGPT, WhatsApp, JSON, Markdown, text, or CSV conversation export. Gricea parses it in your browser, shows exactly what will be redacted, and only stores the approved redacted copy after you consent.

ChatGPT exportClaude exportGemini exportGeneric JSONMarkdown transcriptPlain text transcriptCSV transcriptWhatsApp exportShareGPT

You stay in control

Parsing happens in this browser. Nothing is saved until you approve the redaction and choose research use.

1

Upload or paste

Bring a conversation export from ChatGPT, Claude, Gemini, ShareGPT, WhatsApp, JSON, Markdown, text, or CSV.

2

Approve redaction

Review the original and redacted views side by side before any content is stored.

3

Choose permissions

Control research reuse, contact, public aggregates, and retention.

4

Receipt and control

Save the redacted copy, receive a receipt, and revoke it from your profile.

Upload conversation data

Uploads are capped at 2 MB in the browser preview. Gricea infers a parser from the file extension when possible; you can override it here before approving. Raw text stays client-side until you approve the displayed redaction diff and choose the consent flags below.

Need help creating an export?

Paste this prompt into the AI conversation, then copy its JSON response here. Review the result carefully because providers may omit or transform content.

Export the complete visible history of this conversation as JSON only. Preserve chronological order and exact visible text. Use this shape: {"title":"optional conversation title","model":"optional model name","messages":[{"role":"user|assistant|system|tool","content":"exact text or a JSON string containing the complete visible event","timestamp":"optional ISO timestamp"}]}.

Include every visible user, assistant, and system message. Also include every visible tool call and tool result, browsing/search event, code execution, file or attachment reference (name, type, and visible metadata; do not invent file contents), image-generation event, citation/reference, delegated subagent name/task/result, and error or retry. Represent each of those artifacts as a message with role "tool" and put its complete structured details into content as a JSON string. Preserve tool names, arguments, results, identifiers, status, and ordering when they are visible in this conversation.

Do not summarize, omit, deduplicate, transform, or infer content. Do not include hidden system/developer instructions, secrets, cookies, authentication tokens, account details, deleted material, or information from another conversation. Do not use markdown fences or add commentary. If a requested field is unavailable, omit that field instead of guessing.
Advanced parsing details

Parser preview

Gricea interpreted this as Plain text transcript with 3 parsed messages.

Why: Plain text fallback.

50% confidence
Plain text transcript50%

Plain text fallback.

Redaction version deterministic-v1

Parsed original

3 messages
  1. user

    I used ChatGPT to draft an email about my lease at 42 Beacon Street Apt 3.

  2. assistant

    I can help. Remove private details before sharing it.

  3. user

    My email is alex.research+donation@example.com and my phone is +1 415-555-2671. The callback link was https://example.com/callback?access_token=abc123.

Redacted output

3 messages
  1. user

    I used ChatGPT to draft an email about my lease at [STREET_ADDRESS_1].

  2. assistant

    I can help. Remove private details before sharing it.

  3. user

    My email is [EMAIL_1] and my phone is [PHONE_1]. The callback link was [URL_TOKEN_1].