Recall 2.0 Turns Your Knowledge Base Into a Personal AI

The pitch from Recall is blunt: AI gave everyone the same brain, so your personal knowledge is the only thing left that differentiates you.

That’s a strong claim, and I think it’s mostly right.

Recall 2.0 launched this month and the product is doing something I’ve been waiting to see someone do properly. The original Recall 1.0 was a knowledge-base tool, the kind of thing where you saved articles, notes, and research, and the app would summarize and link them together. Useful, but not dramatically different from a smarter bookmarking service. Version 2.0 turns that stored knowledge into an active AI layer. You’re not just retrieving things you saved. You’re querying an AI that has been grounded in your specific body of work, your notes, your saved articles, your podcast clips, all of it, and you can point that AI at the open internet too, or just keep it inside your own archive. The framing on their Product Hunt page puts it well: “Talk to your knowledge, the internet, or both.”

That toggle matters more than it sounds.

Most AI chat products give you two bad options. You either use a model with a training cutoff that doesn’t know your stuff, or you paste in documents manually every session like it’s 2023. Recall’s approach is closer to what knowledge workers actually need. You build the archive over time, the AI learns the shape of what you’ve saved, and then you ask it things that would be impossible to ask a generic model. “Find the exact clip in my podcast.” “Condense my research on X.” “Compare this new study to the three I saved last quarter.” Those prompts aren’t hypothetical. They’re from the product’s own description, and they show a real use case rather than a demo scenario built to impress.

The user numbers are the first thing I checked. Recall claims 500,000+ professionals on the platform, with logos from Stanford University, NYU, LinkedIn, and Bloomberg visible on the site. I can’t independently verify those logos mean formal institutional partnerships rather than individual users from those organizations, so take the social proof with appropriate skepticism. But 500,000 is a number that, if real, puts this well past “interesting experiment.”

What actually changed between version 1.0 and 2.0 is worth walking through carefully, because the jump is significant. The original product was passive. Save things, get them organized and connected, review them later. It was a smart second brain in the classic sense, something like Notion or Obsidian with better automatic summarization. The new version adds a conversational interface grounded in your saved content, lets you choose which AI model you’re running against, and ships with both an API and MCP support. That last part is meaningful for anyone building on top of their personal knowledge base. The API access means Recall can plug into other tools. The MCP support means it can connect with a broader set of model integrations. This isn’t a closed garden.

The model choice angle is interesting and I don’t think it gets enough credit.

We’ve spent two years watching AI products lock you into one model. You use the app, you use their model, you get what you get. Recall’s approach says you pick the model, which means as better or cheaper models ship, you can swap. For a knowledge base where you’re storing genuinely sensitive professional research, that also matters for trust. You’re not being told your data is going to one specific foundation model with one specific training policy.

I want to be direct about what I can’t assess from the outside. The quality of the AI’s grounding in your specific saved content is the whole product. If the retrieval is mediocre, if the model keeps hallucinating details from your own notes, the value proposition collapses immediately. Nothing in the public materials tells you how well the grounding actually works on a corpus of, say, 3,000 saved articles and 200 personal notes accumulated over two years. That’s a non-trivial retrieval and synthesis problem. I’d want to run it for 30 days with a real research workload before I’d tell you it works.

The “pick a movie based on what I love” example from their own copy is either charming or concerning depending on your read.

On one hand, it shows the flexibility of the system. If you’ve saved movie reviews, letterboxd-style notes, and opinions on what you’ve watched, the AI can draw on that to make a recommendation that a generic model couldn’t. That’s a real differentiator. On the other hand, it’s a strange thing to lead with when your primary market is Stanford researchers and Bloomberg professionals. It suggests the team is still finding the right register for who this is for.

The Android listing in the product categories, alongside Productivity, Notes, and AI, suggests mobile is part of the experience. A knowledge-base tool that only works on desktop has a real limitation. You save things on your phone constantly, links from newsletters, voice notes between meetings, screenshots of things you want to read. If Recall has a solid mobile experience, that’s an underrated part of the product.

Worth putting this in the broader context of where personal knowledge management tools are right now. The PKM category had a massive wave of interest between 2020 and 2023. Tools like Roam Research built cult followings. Obsidian grew a plugin community that rivals some professional software. Notion became ubiquitous. But most of those tools are fundamentally storage and organization systems. The AI integration has been bolted on unevenly. What Recall is doing with 2.0 is trying to make the AI layer the primary interface, not a secondary feature you invoke occasionally. That’s the meaningful architectural bet. According to the PKM research community at NYU’s information science department, the move from retrieval to synthesis is where most personal knowledge systems have historically stalled out.

“The edge is your knowledge,” the team said in the product description.

I believe the thesis. The part I’m watching is execution. The gap between “AI grounded in your knowledge” as a concept and “AI that reliably synthesizes your 1,500 saved articles without losing the thread” as a working product is where tools like this either justify themselves or disappoint. Recall 1.0 reportedly built real retention among professionals, which is more than most PKM tools manage. That gives them a foundation. Whether 2.0 turns the stored knowledge into something that actually changes how you work is the question the next few months will answer.

It did well on launch day, landing at #7 on Product Hunt’s daily rankings, which suggests the existing user base cares about where this is going.

Free to start, API included, your model choice. For anyone sitting on a knowledge base they’ve built over years and never quite figured out how to use: this is probably worth an afternoon.