Launched this week
Memoriq
Your private AI memory for ChatGPT, Claude, Gemini and Grok
196 followers
Your private AI memory for ChatGPT, Claude, Gemini and Grok
196 followers
Memoriq is your private AI memory for ChatGPT, Claude, Gemini and Grok. Save the conversations that matter in an end-to-end encrypted vault that only you can access. Open source, self-hostable, and built for people who don't want to lose valuable AI chats or trust another plaintext cloud service. Search, organize, and keep your AI knowledge under your control.
Multiavatar in Solidity
the multi-platform support across ChatGPT, Claude, Gemini, and Grok is the right coverage but the capture mechanism for each is probably different. some of these have APIs, some require browser extension scraping, and the reliability and permissions vary a lot between them. curious how the actual capture works across platforms and whether there are any that work more reliably than others or require different setup steps
Multiavatar in Solidity
@ansari_adin Good question!
The capture path is provider-specific. ChatGPT is currently the most reliable because the extension uses OpenAI's same-origin conversation API while you're logged in, so it can retrieve the conversation directly from the provider.
Claude, Gemini, and Grok currently rely on DOM extraction, since the goal is to work with the normal web interfaces rather than require separate API keys. Those providers change their UIs fairly often, so capture is best-effort. Gemini, for example, may require manual scrolling to the top for very long chats.
Since browser extensions are generally not available on mobile, the long-term vision is to support native sharing on mobile, whether through the Memoriq PWA or a dedicated app, so useful AI conversations can be saved as naturally as they are on desktop.
@giekaton that's a clear breakdown. the ChatGPT reliability advantage makes sense given same-origin access. the DOM extraction for the others being best-effort is a fair tradeoff given the goal of not requiring API keys. good luck with the mobile path
The capture asymmetry between platforms is the interesting engineering constraint here. ChatGPT via API is reliable, but Claude, Gemini, and Grok depend on DOM extraction. Meaning a UI change on any of those platforms can silently break capture or produce malformed saves.
How do you handle version drift detection? Is it a user-visible failure (capture stops until an extension update ships) or does Memoriq degrade gracefully and capture partial content? Claude in particular has changed its conversation UI structure multiple times this year. What does the extension update cadence look like when a platform breaks?
Multiavatar in Solidity
@binu_george That's one of the main engineering challenges. Right now, if a provider changes its UI, capture quality can degrade until the extension is updated, although partial captures may still work depending on the change. Keeping up with provider changes is an ongoing part of the project.
End-to-end encrypted + self-hostable is the combo that makes this trustworthy for anything work-related. The moment your AI history sits in someone else's plaintext cloud, you can't put real client or business context in it. Open source seals the deal because "private" only means something when you can verify it. The capture across four assistants is the hard part though - they don't all expose the same hooks. Curious how you handle that without it breaking every time one of them changes their UI.
Multiavatar in Solidity
@david_marko Thanks! The capture path is provider-specific. ChatGPT is the most reliable because it can use the provider's conversation API, while Claude, Gemini, and Grok currently rely on DOM extraction. UI changes do happen, so the extension is designed to evolve alongside the providers and improve over time.
The user-curated memory choice makes sense for launch and support workflows where only a few AI conversations are worth keeping as project history. The first thing I would test is whether a teammate can save a Claude/ChatGPT thread, search it later, and prove it stayed private without turning it into another shared knowledge base. Does Memoriq support any team handoff model yet, or is the intended boundary strictly one private vault per person?
Multiavatar in Solidity
@hazy0 Right now, the focus is one private vault per person. End-to-end encrypted sharing is something I'd love to add in the future, and I think it could work well for teams.
@giekaton That boundary makes sense for beta; shared vaults would add a lot of permission surface fast. For team use, even a simple export/import handoff with the privacy state visible could cover the first support playbook workflow before full shared vaults.
Multiavatar in Solidity
@hazy0 That's a good point. There's probably a nice middle ground between personal vaults and full team collaboration.
Exactly. A handoff pack feels like that middle ground: export the selected threads, keep the encryption/privacy state obvious, and let the receiving teammate import it into their own vault. That would cover support playbooks without turning Memoriq into a shared workspace too early.
Ok but what can I do with stored conversations? Can I search across all my saved chats? Can I use them in other LLMs?
Multiavatar in Solidity
@zopteraa Yes, you can already search across conversations saved from different AI providers in one place.
Right now, exports are encrypted vault backups rather than a format for other LLMs. Memoriq is still in beta, and I'm focused on getting the core experience right first.
Longer term, I think there's a lot of potential for richer semantic search, privacy-preserving AI agents that can work with your own conversation library, end-to-end encrypted sharing, and new ways to connect knowledge across different AI providers.
Congrats on the launch! The E2E encryption and self-hosting options are a huge win for anyone worried about data privacy with LLMs. What cryptographic primitives are you using to handle the end-to-end encryption without bottlenecking the search/retrieval speed?
Multiavatar in Solidity
@luyanda_ntombela Thanks for the question!
The current approach is to keep the cryptography fairly conventional and avoid clever encrypted indexing schemes for now. Memoriq uses a browser-generated 256-bit AES-GCM master encryption key (MEK). The user's encryption password is processed with PBKDF2-SHA256, the MEK is wrapped with AES-KW, and conversations are encrypted locally before upload.
To avoid search becoming a bottleneck, the server doesn't perform content search at all. After you unlock the vault, encrypted conversation headers are decrypted locally in the browser and used for list and search operations, while full conversation bodies are decrypted on demand. Server-side semantic search is something I'd like to explore in the future, but only if it can preserve the privacy model.
There's a more detailed write-up of the encryption and extension architecture in the repo as well, if you're interested in the implementation details, you can find it on GitHub.