Somewhere in a Slack thread three months ago, an engineer mentioned why the team killed that feature. The decision got documented in a Notion page. Or maybe it was Google Drive. Either way, when someone asks ChatGPT or Claude about it today, the AI has no clue.
This gap—between what happened inside a company and what its AI tools can access—is what Shalin Shah and Kanyes Thaker are trying to close. Their startup, Hyper, emerged from Y Combinator's Spring 2026 batch with a pitch that sounds almost too simple: what if AI assistants just knew things about your company, automatically, without anyone having to explain?
The two founders, who previously worked together at robotics company Matic, went public with their product in late May. The concept isn't a new chatbot or another enterprise search tool. It's infrastructure—a self-maintaining knowledge graph that ingests data from Slack, Gmail, Notion, GitHub, Google Drive, and calendars, then feeds relevant context back into AI assistants as people use them. No manual curation required. No separate dashboard to check.
Whether this actually works at scale is, of course, the open question.
Automated Context, Theoretically
The technical premise revolves around what Shah and Thaker describe as a knowledge graph that maintains itself. Not just indexing documents, but deduplicating information, resolving conflicts when the same fact appears in multiple places, and updating continuously as new data flows through a company's tools.
Shah explained the retroactive piece during Hyper's Product Hunt launch: the system "goes back and indexes Slack history (and … Gmail, Drive, Notion, etc.)" from the beginning, then keeps up with live updates. The idea is that when you're chatting with Claude or Cursor, Hyper injects the relevant background information directly—no need for what the founders dismissively call "ugly tool calls" or manual context-setting.
Third-party reviewers have suggested Hyper uses something called Model Context Protocol servers to handle this injection, though the company's own materials haven't explicitly confirmed that architecture. What's clearer is the ambition: to make AI tools smarter without changing how anyone uses them. Background intelligence, essentially invisible.
For a two-person team building in San Francisco on Y Combinator's standard $500,000 deal ($125,000 for 7% equity, with the remainder via an uncapped MFN SAFE), that's a significant engineering challenge. Especially if you're promising it will work reliably across hundreds of data sources.
Demand, If Not Proof
Timing may be on their side. Y Combinator added "Company Brain" to its official Request for Startups list earlier this year—a clear signal that the accelerator sees shared memory infrastructure as increasingly critical as companies deploy more AI agents that need to coordinate.
Hyper's launch seemed to validate that instinct, at least initially. The startup hit Product Hunt on May 29, landing at #12 for the day. A Hacker News launch followed on June 3. Within less than two weeks, according to the company's Y Combinator profile, they'd onboarded over 50 teams and reached $1,000 in monthly recurring revenue. Modest, certainly. But for a product that had just gone live, it suggested people were feeling the pain.
Cofounder Thaker put a finer point on it in a LinkedIn post, claiming 500+ daily active users immediately after launch. The team has also mentioned scoping paid design pilots with companies including Razorpay and Snorkel AI, though these remain company-reported engagements.
Still—50 teams and $1,000 MRR is a start, not a victory. Plenty of early-stage products attract curious early adopters. The harder part is proving the system actually delivers on the promise when it's managing thousands of documents across dozens of tools.
A Crowded Field, Competing Theses
Hyper isn't the only startup chasing this idea. Y Combinator's Spring 2026 batch alone includes at least two others in similar territory: Hyperspell, positioning itself as a "permission-aware unified source of truth," and Memory Store, which launched in May as a "shared company brain for teams and AI agents."
The terminology is almost interchangeable. The architectures, presumably, are not.
And then there's Garry Tan. In April, the Y Combinator CEO open-sourced GBrain—his own take on agent memory systems. The code drop sparked a fair amount of discussion in developer communities about how these architectures should be built, which probably didn't hurt the case for funding startups in this category.
The established players—Glean, Notion, Guru—already serve parts of this market, but they typically require manual curation or force teams to use separate interfaces. Hyper's bet is that automation and invisibility matter more than another tool to log into. Whether that's the right bet depends on whether the automation actually works.
The Messy Bits
Some practical questions haven't been resolved yet. When someone asked about GDPR compliance during the Product Hunt launch, Shah's response was candid: "Not yet but definitely in the works!" A review from May noted the absence of public compliance documentation—something that matters quite a bit if you're asking companies to feed internal data into your system.
Pricing remains fuzzy. Multiple reviewers describe the product as "free during early access," with official pricing still to be announced. One Spanish blog mentioned a three-day trial, though that detail hasn't appeared anywhere else.
There's also ambiguity around exactly what's shipping versus what's planned. The technical claims about Model Context Protocol integration and references to 160+ connectors come primarily from third-party reviews rather than the company's own materials. That could mean the features are live but undocumented, or it could mean observers are extrapolating from roadmap talk.
Either way, it leaves some gaps in the public narrative.
What Comes Next
For Shah and Thaker, the challenge now is execution at a scale that hasn't been tested yet. The self-maintaining graph thesis assumes that automated context injection will work reliably across hundreds of data sources and dozens of AI tools. That's a significant technical lift, particularly for a team of two.
If the founders are right about market timing—that the infrastructure layer for AI coordination is about to become critically important—then the next few months matter. As more companies deploy agents that need to share context and coordinate actions, the systems that enable that coordination become increasingly valuable.
Whether Hyper's particular approach proves to be the right architecture is still an open question. But the early demand, however modest, suggests they've identified something real. Now comes the harder part: delivering on it.
