Memory Store (YC Spring 2026): teardown vs XMarts

In one line: Memory Store sells a shared memory that the AI agents of a small team write to and read from, over MCP, at a price per person [site] [yc]. Threat: 2, because it is the same "company brain" idea for the same kind of small team and it is YC funded, but it is early, has no named customer and stops at one layer [site] [press].

1. Fact file

2. Their proposition, reverse engineered

3. Positioning

• Category they claim: "Company Brain" on the YC page [yc]. On the live page the claim is smaller: team memory for agents, with a "company brain" listed as the thing they are building toward [site].
• Category a YC partner would put them in: shared agent memory for small AI-native teams (inference).
• What they want to be known for: the memory layer every team's agents read from and write to.
• Where they sit on the map: L1/L2/L3 = 3/1/0. Buyer: sme, developer-led (inference).

4. Three layers vs ours

5. Same and different

6. Wedge overlap

7. Whitespace (where they stop)

• Layer gap: memory only. They do not run, check or measure work [site]. To own it we need one workflow where memory feeds agents that finish work and a checked number moves.
• Buyer gap: the site names no sector and no country, and shows no customers [site]. UK consultancies and agencies are not addressed. To own it we need calls with named people at those firms who say they have the pain.
• Number gap: no result, no usage figure [site] [yc]. To own it we need one before/after number from a live customer workflow.
• Price gap: $150 per person per month, and read-only users are billed [site]. A 30-person firm would pay about $4,500 a month (arithmetic on their price). To own it we need a price tied to the result, not to seats.
• Speed gap: not a gap. They offer a 7-day trial and a three-step setup [site]. We would need first value in days to match them.

8. For the YC application

• Proven today: Marton's logs show his system keeps a company memory (about 1,400 pages, 112 call transcripts, about 8,700 quotes) and that the same system runs about 5,600 agent jobs, each checked by a different AI, with about 160 failure records that each added a check. Memory Store's public product stops at memory [site].
• Draft answer to "how are you different from Memory Store?": DRAFT. "Memory Store stores what a team's agents remember. Our system already uses its memory to run agents that do the work and check it, every day, on our own business." Founders say it their own way.
• Future claim (needs proof): "Our memory improves because every result is checked, and one workflow for a UK firm gets measurably better." Needs one customer workflow run for 4+ weeks with a before/after number and a rule-change log tied to checked runs.
• What we understand that they don't: a memory is only as good as the checks on what goes into it, and you need the work loop to get those checks. needs proof with a customer; Marton's logs show it on his own work only.
• Copyability (M3): medium. Evidence: their roadmap says they will add tools and flag overdue commitments and decisions that affect people [site], which is a step toward acting, so the idea is in reach. But running and checking agents is a different product, and they only have 4 people and sell per seat to agent users (inference). Not hard, so M3 can not score above 3 against Memory Store alone.
• Why now (M4): None found in this teardown. MCP adoption is general AI progress, not a dated change tied to our edge.
• Do not claim: "company brain" as our headline (their YC line uses it and a YC partner already funded them). Anything about their product being weak on capture quality (not tested). "No one does the checked loop" (UNVERIFIED across the field; only this company was checked). A buyer count or customer for our system (none yet).

9. Ideas worth taking

1. Three-verb agent contract: checkin, record, recall. Small and easy to explain in one breath [site].
2. A personal morning email that says what changed, what is blocked and who waits on you. A buyer can feel this on day one [site].
3. Open the product to a stranger in a 7-day trial with a three-step setup. Fast first value, no sales call needed [site].

10. Unknowns

• Real customers, usage and revenue. The homepage and YC page show none. Look for a Launch YC comment thread or founder posts on X.
• When the rename to MemStore happened and why the live page is narrower than the YC pitch. The Wayback Machine could not be fetched here; check it by hand.
• Funding amount. Directory figures conflict ($500K versus undisclosed).
• Whether they have a security page, retention policy or SOC 2. The site names none, and its own sample email lists the questions as open.
• Whether Briefs and Slack auto-capture ship today or are roadmap. The YC page says yes; the live page says Slack decisions must be recorded by you or your agent.

Sources

• mem.store (memory.store redirects here) [site], checked 2026-10-19
• YC company page [yc]
• YC Launch post: Memory Store, your company's brain [yc], undated ("5 months ago" on the page)
• VCBacked profile [press], June 2026 data
• YC Tier List review [press]
• Julep docs [docs], founders' earlier product
• Composio post by Diwank Tomer, Julep AI [blog]