← Library

FAQ — privacy, features & supported files

Straight answers. Your data is private by default and lives in your browser. Here's exactly what we do, what we don't, and what we can unpack for you.

Privacy & security

›Can you (the developer) see my chats or uploaded data?
No. Your conversations, uploads, folders, tags, favorites, and notes are stored locally in your browser (IndexedDB) on the device you use. We do not have a server-side copy of your chat archive and cannot browse it, export it, or hand it to anyone. Clearing your browser storage or switching devices wipes the library — that is the trade-off for keeping it private by default.
›What is sent off my device, ever?
Only what you explicitly trigger:
  • AI features (Ask, Interpret, Surprise Me, Reports, Patterns): the specific slice of conversation text needed to answer your prompt is sent to our AI gateway to run the model. We do not store it as part of your account.
  • Account & sign-in: email/auth handled by our backend provider (industry-standard hashed credentials, RLS, HTTPS).
  • Error logs: if something crashes, a short stack-trace + page URL is logged so we can fix bugs. No chat content, no file contents.
  • Feedback you submit via the Feedback page.
Uploading a file does not upload it to us — parsing runs in your browser.
›What standard security measures are in place today?
  • HTTPS/TLS on every request.
  • Authentication + Row-Level Security on the backend, so the only rows your session can read are your own.
  • Roles stored in a separate table with a security-definer check — no client-side admin flags to tamper with.
  • Publishable/anon keys only in the client; secret keys stay server-side.
  • Your chat archive is browser-local (IndexedDB), scoped to this origin — other sites can't read it.
  • Automatic security scans on the backend after schema changes.
›What's coming next on security & privacy?
  • Optional end-to-end encrypted cloud backup of your library.
  • One-click "wipe everything on this device".
  • Opt-in redaction pass before any AI call (names, emails, keys).
  • Exportable audit log of every AI call you made.
›Being honest — what are the limits?
Browser-local storage means a device wipe = data gone; back up your source export files. AI calls go through a third-party model provider under their terms — we don't train on your data, but we can't cryptographically prove that end-to-end today (see roadmap above). If regulated data (PHI, financial identifiers) is a concern, wait for the redaction release or don't run AI features on it.

What CloudKey does today

›Library & organization
  • Unified library of every conversation across providers.
  • Folders, tags, favorites/bookmarks, custom filters.
  • Full-text + keyword + semantic (embedding) search.
  • List / rows / grid views, sort by recency, length, relevance.
›AI-powered analysis
  • Ask — natural-language Q&A over your archive.
  • Interpret — turns fuzzy queries into structured filters.
  • Surprise Me — resurfaces forgotten but interesting threads.
  • Reports & Patterns — themes, topics, prompt habits over time.
  • Prompts library — save, tag, and reuse your best prompts.
›Repackage & export
  • Clean Markdown export per conversation or per folder.
  • Shareable, printable reading views.
  • Blog-ready formatting (see the "How to export AI history" post).
  • JSON round-trip so your data stays yours.
›Account & workspace
  • Email + Google sign-in.
  • Feedback channel wired directly to the team.
  • Plans page for upgrades and Donate for supporters.

Files we can unpack, analyze & repackage

›ChatGPT — official export
Drop in your conversations.json (or the whole export zip's JSON). We walk the full mapping tree — parent/child branches, tool messages, multi-part content — and rebuild each thread in order.
›Claude / Anthropic — official export
data-*.json or the conversations.json from the Claude export. We parse the chat_messages array, preserve roles, timestamps, and attachments-as-text.
›Generic JSON chat dumps
If a file isn't recognized as ChatGPT or Claude shape, we fall back to a generic parser that looks for messages[] with role + content — works for many Gemini/Copilot/OpenRouter/self-hosted exports.
›Plain text & markdown
.txt, .md, or any text file becomes a single-message conversation you can tag, search, and feed to the AI tools alongside the rest of your library.
›What about PDFs, DOCX, images, audio?
Not yet in the client-side parser. On the roadmap: PDF & DOCX text extraction, ZIP auto-unpack, image OCR, and audio transcription — all running locally where possible. If you need one of these sooner, tell us on the Feedback page.

Something missing or unclear? Tell us on the Feedback page — we keep this FAQ honest.