Hyper (YC Spring 2026): teardown vs XMarts

In one line: Hyper reads a small team's docs, Slack, email and calendar, builds a shared "company brain", and feeds it to the AI tools the team already uses [yc] [launch]. Threat: 2, because it targets the same small-team buyer and our L1, but it is two people, three months old, and stops at memory and light task help [yc] [docs].

1. Fact file

2. Their proposition, reverse engineered

3. Positioning

• Category they claim: "Self-driving company brain" [site]. Company brain for AI employees [yc].
• Category a YC partner would put them in: shared agent memory and context for small AI-native teams (inference).
• What they want to be known for: the context layer your AI tools read without you doing anything.
• Where they sit on the map: L1/L2/L3 = 3/1/1. Buyer: sme, business (with a developer-tool entry through coding-agent hooks).

4. Three layers vs ours

5. Same and different

6. Wedge overlap

W9 (ask my agent before you book my time) is a distant neighbour: shared knowledge lets a teammate "pick up work a teammate started" [docs]. Not scored here.

7. Whitespace (where they stop)

• Layer gap: they supply context; the work happens in someone else's agent, or in a light app [launch] [docs]. To own it we need one customer workflow where our agents do the work, a different AI checks it, and the brain updates from the result, run for 4+ weeks.
• Buyer gap: they skip regulated and government users (founders' own words) and sit in tech and AI-native teams [hn]. Services firms and consultancies are not named [yc] [hn]. To own it we need services-firm buyers who say they have the call-memory pain (W3 has no calls yet).
• Number gap: no benchmark scores or customer result published; the traction is signups and $1K MRR [hn] [yc]. To own it we need one before/after number a buyer would sign.
• Price gap: not testable. Price is not public. Unknown.
• Speed gap: not a gap. They claim setup under 20 minutes [launch].

8. For the YC application

• Proven today: Marton's logs show a company brain that is real data (about 1,400 vault pages, 112 call transcripts, about 8,700 customer quotes) and agents that act on it: about 5,600 jobs, each checked by a different AI, with about 160 failure records that each added an automatic check. Hyper's public results are signups and revenue, with no work-done or benchmark numbers yet [yc] [hn].
• Draft answer to "how are you different from Hyper?": DRAFT. "Hyper gives your AI tools better context. Our system also does the work, has a different AI check it, and feeds the result back into the brain. We run that on our own business every day." Founders say it their own way.
• Future claim (needs proof): "Our agents do the work for a services firm and the brain gets better from each checked result." Needs one customer workflow run for 4+ weeks with a before/after number.
• What we understand that they don't: a brain improves faster from checked results of finished work than from reading sources alone. needs proof on a customer; Marton's logs show it on Marton's own work (offer feedback lines fell from 44 to 18 as rules were added).
• Copyability (M3): medium. They already extract facts from each agent response and accept human corrections [hn], so adding a result-check loop is a natural next step and a team of 2 can ship it in months (inference). But a check-and-measure loop needs a customer workflow and an owned outcome, which pulls them from a self-serve context product toward doing the work (inference). Not hard, so M3 can not score above 3 against Hyper alone.
• Why now (M4): None found in this teardown. Do not use "agents can now do the work" or "agent memory is hot": that is general AI progress.
• Do not claim: "company brain" as our headline (their YC line and site use it). "We have customers" (none yet). "They don't do the work" without the caveat that their desktop app is growing [docs]. That the brain is multi-user with permissions (Marton's logs show it is single-user). Anything about their funding or customers (UNVERIFIED).

9. Ideas worth taking

1. Keep old facts and mark them superseded, instead of deleting them. It lets a user see why an answer changed [hn].
2. Push context in through coding-agent hooks, which they say work better than tool calls [hn]. Ideas only.
3. A one-line /correct fix path so humans repair the brain in the middle of work [hn].

10. Unknowns

• Their price and whether the 3-day trial turns into a paid plan. Look at the app's sign-up flow or ask a user.
• Whether the 50+ teams and $1K MRR are real and still growing, and whether Razorpay or Snorkel signed. Re-check their X account and the YC page later.
• Funding amount and investors. Only aggregators list it.
• Benchmark scores (LoCoMo, LongMemEval). They promised to publish [hn]; check their site and blog.
• Whether the desktop app now does multi-step agent work (would raise L2/L3). Read the docs release notes.
• UK or EU customers. GDPR support was "in the works" in the Product Hunt replies [producthunt]. The marketing site is a script-built page and the fetch returned only the headline, so site claims came from the YC, launch and Product Hunt pages.

Sources

• [site] heyhyper.ai, headline only readable
• [yc] YC company page
• [launch] Launch YC: Hyper, the Self-Driving Company Brain
• [hn] Launch HN thread, read through a summary
• [producthunt] Product Hunt: Hyper, 29 May 2026
• [docs] Hyper docs, Hyper 1.11.0, nightly 18 Oct 2026
• [press] Ecosistema Startup writeup (trial, integrations)
• [press] FundedIQ and Neuronfeed, funding figures, low confidence
• [jobs] YC company page lists 0 open jobs