Memdex
Turn every AI conversation into reusable local memory
Start new thread

Memdex - Turn every AI conversation into reusable local memory

by•
Memdex is a Chrome extension that turns every AI conversation into reusable memory. It auto-captures your chats across ChatGPT, Claude, and Gemini, stores them encrypted on your laptop (IndexedDB, never uploaded), and underlines reusable context in your next prompt like Grammarly underlines typos. One click to inject when you need it just like grammarly. Free for your 10 most recent chats. Pro for unlimited.

Add a comment

Replies

Best
Maker
📌
Hey Product Hunt 👋 I'm Andrew. After spending most of last year jumping between ChatGPT, Claude, and Gemini, three things wore me down: 1. Every new chat starts at zero. I keep re-explaining the same project, the same constraints, the same voice. 2. My "memory" lives in someone else's cloud — locked to one model, invisible, unmovable. 3. The existing fixes are heavier than the problem: vector DBs, accounts, settings panels, prompts called /remember. So I built Memdex. Memdex is a browser extension that does three things, automatically: • Saves every chat on ChatGPT, Claude, Gemini, Perplexity, and Grok to encrypted files on your laptop. Never uploaded, never trained on. • Detects when something you already discussed is relevant to what you're typing now — like Grammarly, but for context. • Injects that memory into any AI tool with one click. No copy-paste. No re-explaining. The aha is the second tab: you open a fresh Claude window and it already knows your project, because last week's ChatGPT thread is on your disk and Memdex just pulled it in. It's local-first (memory never leaves your machine), cross-model (one memory, every tool), and zero-ritual (no /remember, no DB, no signup to start). Free tier auto-saves your 10 most recent chats; Pro is unlimited. 👉 Add to Chrome: ... 👉 Site: What I'd love feedback on: - Where "detect" feels noisy vs. genuinely useful - Which tools you want next (Cursor, Windsurf, v0, Lovable are on the list) - Anything that makes you not believe "local-first" — I want to fix the perception, not just the reality Building in public from here — happy to answer anything in the thread. — Andrew

 I use Claude with a browser automation extension to handle long tasks. The problem is the context window fills up mid-task, and I have to star a new conversation — but the job isn't done yet. Every time I restart, I have to re-explaiin everything from scratch: what I was doing, how far I got, what's next.

Does Memdex handle this kind of scenario? Can it capture and resume that kind of task context, not just regular chat conversations?

Maker
Thanks for your interest, and we do support this kind of scenario. You can just click the Memdex icon in the other chatbox that you startover again and append the past conversation in it

 "never uploaded" and "injected into Claude on every relevant prompt" are both on your page, and both true. that's the part I'd want addressed head-on, Andrew.

the memory rests on my machine, sure. but the second it's useful, it travels into someone else's model. local-first storage, cloud-first exposure.

do users get to mark certain memories as never-inject? the work-stuff-yes, the therapy-session-no kind of split? feels like that toggle is what separates "encrypted on disk" from actually private.

congrats on the launch either way, this is a real problem worth getting right.

 Love the idea of cross-model memory. One thing I’m curious about: how are you handling relevance detection locally without turning into “prompt bloat”? Like, what signals decide which past chats get injected vs ignored?

 I use claude pro and store all my files in a project and have conversations using those project repositories. The issue i face is sometimes chat conversations get long and once it exceeds a limit, claude asks to open a new chat . When I open a new chat, I got to ask it to summarize previous chat information and use it back and it doesn't fully recognise the conversation style or the details to full extent. Does this address the scenario?

I just woke up and opened Product Hunt to see what’s launching today. Memdex caught my eye so I went through the homepage, and the way you described the context detection flow felt clean. Great work! By the way I lose my best AI chats all the time. Scrolling through the history is a nightmare. So this idea... I like it. I have a question though. Since everything stays on my device, how does the matching actually work? Is it just looking for the same words, or does it understand the meaning behind them? Either way, cool product. Rooting for you.

Maker
Hey there, thank you for supporting us! Our product basically records every conversation locally. Then, for specific projects, we summarize and learn from those conversations — your context, preferences, and working style — and index them into project memory. Everything happens behind the scenes. When you prompt anywhere, Memdex can inject the relevant markdown file, just like Grammarly. The magic just happens in the background.

 Thank you for the detailed response. So it learns my style over time. That's smart. I will keep an eye on how this grows. One more thing... do you have a LinkedIn page or Twitter for Memdex?

Hey Andrew! Sounds obvious but it actually isn't. Also jumped between all of them and know how hard is to set again the context. I'm sure ti's gonna change the game. Wish you all the best!

I literally had a discussion with Claude about this and few days ago and wanted to build something. No the issue remains: how do I get my mobile conversations tracked as well?
Maker
Nico, thank you for your question.I'm the short answer is we could not inject your memory in mobile version because we are in the Chrome extension. But once you open the Chrome extension, we could reload the past session history that you chat in the mobile phone, and save that locally in memdex
chrome extensions work on some mobiles as well afaik?

"Reusable local memory" is the part most AI tools paper over with "your data is private" wording without actually solving the locality problem. Curious which decision was harder — the storage shape or deciding what NOT to capture. That second one usually defines how a memory tool ages.

Maker
Hey, thank you for your great question! This was one of the hardest parts for us too. We store the full conversations locally, then do lightweight indexing and episodic memory extraction per project. For aging, we separate things into long-term memory, project memory, and changelog/source history, instead of surfacing everything forever. Totally agree that what NOT to capture is what makes memory age well. Would love to learn more about your use case too.
Interesting product. One thing I’m curious about from a technical standpoint though, I’m guessing Memdex relies on scraping data from ChatGPT/Claude/Gemini web apps, doesn’t that become fragile over time? I built a small chrome extension for myself a while ago that scraped linkedin job descriptions, and it kept breaking whenever linkedin changed their dom structure. From what I have seen, a lot of companies intentionally change frontend structures specifically to make scraping harder. Did you think about this problem while building Memdex? Curious how you’re making the integrations stable long-term.
Maker
Yes, that's definitely a time consuming thing, and we do need to keep our eye on the any updates of OpenAI, Claude and Gemini frontend

Reusable conversational memory is one of the biggest unlocks for practical AI workflows.

The idea of memory is extremely helpful. I use many AIs, yet heavy on Claude. instead of copying and pasting prompts or ask it to summarize it then paste it in a different AI, phew!!! this alone is brilliant "you open a fresh Claude window and it already knows your project".

Agent Memory is critical, congrats on the launch!

The Grammarly-for-context framing is strong. The place I’d be careful is separating durable memory from stale task state.

Some things really should follow me everywhere: product constraints, preferred wording, customer language, recurring project facts. Other things are only useful for this week’s thread and become confusing later. I’d love a lightweight way to mark injected context as “keep reusing this” vs “only for this task/session,” plus a small preview of why Memdex thinks a prior chat is relevant before it inserts it.

  Good question, I was wondering the same thing. The "keep this forever" vs "only for this task" distinction is exactly the kind of UX detail that makes or breaks a memory tool long-term. Curious if the maker will chime in on this — a simple tagging or expiry mechanism would go a long way.

Maker
Jim, that's a very important question and our solution on that is to maintain the long term memory file just like what OpenClaw is doing like soul.md, Index them into the project category as well. so when you want to continue your past conversation across different model, we just inject the recent and relavant few section of it, but not sending all of the noise into the model.

That makes sense. The part I’d keep making visible is the “why this section, not the rest?” decision.

If the user can see a small injection preview — project memory, last touched date, 3 snippets selected, 40 other notes ignored — they can trust the context without reading an audit log. It also gives them a chance to catch the dangerous case where old context is technically relevant but no longer true.

12
Next
Last