Augmented Browsing: From Resurfacing to Active Reasoning
P
Paul Richards
Right now, Augmented Browsing resurfaces related content from your knowledge base as you browse. It's a great starting point, but it's mostly a read-only experience. This request is about evolving it into an active reasoning layer that changes how you see the web based on what you already know.
- Active Bridge (Web <-> KB)
Let users highlight any sentence on a page and "Connect to Card," linking a snippet directly to an existing card through a quick search popup (similar to an @-mention flow). This turns browsing into a two-way interaction: instead of just surfacing what you saved, you can build your knowledge graph while reading, without leaving the page.
- Synthesis Overlay (Reasoning over Reminding)
Go beyond highlighting names and entities. Highlight concepts. A keyboard shortcut (e.g. Cmd+K) could let you ask things like "How does this article fit into my Business / Strategy branch?" Recall would highlight the relevant parts, visually map them to your tags, and even nudge you when new content challenges something you saved before.
Why this matters:
This shifts the mental model from "We show you what you saved" to "We change how you see the internet based on what you know." Instead of opening a sidebar to chat with a summary, the knowledge graph would live directly on the page you're reading. It bridges snippets back to your KB and maps concepts in real-time without ever breaking your reading flow. Less "look what I found," more ambient intelligence.
Log In
s
schroell
By contrast, the “Synthesis Overlay” direction is much more interesting to me. Conceptually, this is very close to how I already use Recall together with my "Claude Life" system via MCP: I let an external agent reason over my Recall corpus and new material, mostly in an automated and offline fashion. Bringing that style of reasoning directly into the page – letting me “talk” to my Recall‑curated database in the context of the original article or PDF – would add something genuinely new to my workflow. I can imagine this being especially valuable for deep research PDFs where I have the intuition “I’ve seen similar ideas before, but the details differ,” and I want the system to surface both overlaps and tensions between the new document and my existing notes before I invest more time.
From a product positioning standpoint, I think this second piece is also a strong differentiator. Obsidian has no native AI and relies on plugins that, in my experience, are not particularly convincing. Notion does more out of the box but is, in my view, too expensive for the marginal gain over an Obsidian‑style stack. If Recall can provide a first‑class, in‑page reasoning layer that understands my corpus and the page I’m on, that pushes it clearly beyond “just another note app” and makes it easier for people to choose Recall as their primary or even only “brain.” My perspective is slightly biased because I personally use Recall more as a “third brain” in a broader toolchain, but for users who adopt Recall as their main or second brain, this kind of smart, ambient synthesis on top of Augmented Browsing feels like a big upgrade and a meaningful market differentiator.
s
schroell
First of all, I really like the overall direction of this proposal. Turning Augmented Browsing from a mostly read‑only resurfacing layer into something that actively reasons over my knowledge graph feels like the right long‑term vision.
For my personal workflow, though, I would probably make less use of the “Active Bridge” part. The main way I use the extension today is to handle complex, long web pages: I want a fast, high‑quality synthesis of the page, then store that as curated context in Recall for future work and reference. I rarely read every line of a page in detail and then actively select individual sentences to connect them to existing cards. I simply don’t have the time to operate at that level of granularity, and I also don’t always have a precise mental index of everything that already exists in my Recall graph on that topic. Because of that, a manual “highlight → connect to card” flow would likely see limited use from me in practice.
There is also a dependency on tagging quality. Right now I rely maybe 95% on automatic tagging in Recall. It works “good enough” for me because I only use tags for a rough classification and occasional cross‑referencing, and I’m willing to live with a fairly messy tag tree that I never fully clean up. But I know there are power users who consider the current auto‑tagging weak enough that they explicitly ask for ways to disable it. In that context, a feature that leans heavily on precise, well‑maintained tags and a perfectly structured graph is less compelling for me personally.