XMarts room

Spring 2026

Savant

Company brain every AI-native enterprise runs on

Visit website

What they do

company brain that captures undocumented procedures and serves them to staff and agents

How close to us

3, close

memory is the core product; serving it to agents, no execution or learning loop stated; enterprise buyer

Wedges

  • W3 Company memory from every call, for consultancies adjacent
  • W8 The other 90% of an AI skill adjacent
L1 Company brain (knows the organisation) 3, strong
L2 Agent team (runs agents, tools, approvals) 1, light
L3 Does and learns (finishes, checks, measures, updates the brain) 0, none

Who buys

Firm size
enterprise
Buyer type
business
Role
operations and AI leads (inference)
UK
not known
Their own description (from YC)

Savant is the company brain every AI-native enterprise runs on. We plug into the tools you already use, capture the written knowledge / undocumented procedures, and serve them to your employees and agents at decision time.

Full teardown

Savant (YC Spring 2026): teardown vs XMarts

In one line: Savant reads a company's existing tools, pulls out the unwritten rules of how work is done, and hands them to staff and AI agents at the moment they decide [site] [yc]. Threat: 2, because the words match ours ("company brain") and a YC partner funded it, but it is two people, has no named customer, and stops at serving context [yc] [site].

1. Fact file

Item Finding Source
What they say Site: "The Operational Memory Layer". YC line: "company brain every AI-native enterprise runs on". [site] [yc]
Founded, team, funding Founded 2026, YC Spring 2026, 2 people, San Francisco, YC partner Harj Taggar. Funding: not stated. A directory shows a $125k round but says its own page is not reliable, so treat as UNVERIFIED. [yc] [press]
Founders Anthony Kim (CEO). His YC bio says he studied Electrical Engineering and Computer Science at Yale. A third-party directory also names Kevin Yu as co-founder and CTO; the YC page we fetched lists only Kim. What YC likely saw: a very young technical team in a hot category (inference, no evidence of prior company). [yc] [press]
Customers None named. No logos, quotes or case studies on the site. The terms say the product is a private beta for "selected design partners". [site]
Price Not public. Terms mention card and bank payments and subscriptions that renew; liability is capped at the greater of six months of fees or $100. That fits small self-serve or pilot deals, not large contracts (inference). No job ads. [site] [jobs]
Path Legal entity is Bledger, Inc. [site]. The YC open list shows former names Bledger and DataGate (no dates found) [yc]. Site changelog and guides pages exist but are empty; last updated 27 May 2026 [site]. [site] [yc]

2. Their proposition, reverse engineered

Part What they do
Who buys YC line says "AI-native enterprise" [yc]. The site names no buyer or size. Integrations (Google Workspace, Office 365, Salesforce, Slack, Zendesk, Intercom) fit scale-ups as well as big firms [site]. Likely an operations or AI lead (inference).
The pain AI "cannot operate a business from raw data alone": the rules, exceptions and judgment calls are in people's heads and threads [site]. Staff waste time searching threads and repeating explanations [site].
The promise Teams find the answer fast. Agents act correctly because they get the policy, the constraint and the edge-case judgment at decision time [site].
How it works Four steps: Connect (read existing tools and databases), Capture (extract procedures, exceptions, judgment), Structure (queryable context, kept current), Serve (agents call Savant when they decide) [site]. The privacy page says connected tools can include messages, documents, tickets and calendar entries [site]. No mention of interviewing staff (UNVERIFIED either way).
The output Queryable "operating context": policies, workflows, constraints, judgment. Example questions: when to approve a refund, which pricing exceptions are safe, what an agent must never do alone [site]. How agents call it (API, MCP) is not stated [site].
Proof style None yet. No numbers, no accuracy, no time saved [site].
Who sells it Book-a-demo form and an early-access form [site]. No sales hires visible [jobs].
Price model Not public. Terms point to subscriptions or "design-partner" arrangements [site].
The story Enemy: search tools, chatbots, wikis, knowledge bases ("not" any of these) [site]. Belief: the firms whose AI understands their business will win [site]. Promise to staff: "Nothing to migrate. No new place for people to live." [site]

3. Positioning

  • Category they claim: Operational memory layer / company brain [site] [yc].
  • Category a YC partner would put them in: context layer for enterprise AI agents, next to the other company-brain companies in the batch (inference).
  • What they want to be known for: the place agents ask before they act.
  • Where they sit on the map: layers L1/L2/L3 = 3 / 1 / 0, buyer = enterprise (YC line, inference for the rest), business.

4. Three layers vs ours

Layer Them (score + evidence) Us (Marton's logs show)
L1 Company brain 3. The memory is the whole product: capture, structure, keep current [site]. A vault of about 1,400 pages, 112 call transcripts and about 8,700 customer quotes, searchable by meaning. It learns from calls and from checked work.
L2 Agent orchestration 1. Agents are customers of the product, not run by it. It serves them context; no approvals, handovers or team routing are described [site]. About 5,600 jobs run by agents from three makers (OpenAI, Anthropic, Google), each checked by a different AI, then automatic tests.
L3 Execution and learning 0. It does no work and checks no result. "Kept current" is claimed with no method [site]. Screen pass said 1; lowered after reading the site. About 160 failure records, each with an automatic check. Offer feedback lines fell from 44 to 18 as 22 rules were added. Only about 15% of jobs reached the main code, and merges were unreliable in October.

5. Same and different

Topic Same or different Them Us
Core belief Same Agents fail without the unwritten rules of the business [site]. Same (brief).
Headline words Same "Company brain" in the YC line [yc]. The brief says lead with the result on one workflow, not the brain.
Where the brain is read Same Served to agents at decision time [site]. Marton's logs show the vault and skills are read by agents before they act.
How it fills Different Reads existing tools and databases, extracts the rules [site]. Marton's logs show it fills from recorded calls and from rules added after each failure.
Who does the work Different The customer's own agents and staff; Savant only informs [site]. Marton's logs show our agents do the work and a second AI checks it, on our own business only.
Buyer Different or unknown "AI-native enterprise" [yc]; nothing else public. Target: SMEs and scale-ups (brief). No customer yet.
Proof Same (both thin) No customer, no number [site]. No outside customer pays; one number is internal (44 to 18 feedback lines).
Team Different Two people, one named CEO [yc]. Two founders, one who builds the system and one who sells (brief).
Price Unknown Not public [site]. Not set yet.

6. Wedge overlap

Wedge Fit Why
W3 Company memory from every call adjacent Both capture how a firm works. They read messages, documents and tickets across tools [site]; W3 learns from recorded calls at services firms (brief). Different source and different buyer.
W8 The other 90% of an AI skill adjacent They claim to keep operating context "continuously current" [site], which is part of skill upkeep. No sign of skills, feedback from use, or retiring dead rules (UNVERIFIED that none exists).

7. Whitespace (where they stop)

  • Layer gap: they serve context and stop; they do not run the work or check it [site]. To own it we need one workflow run by our agents for 4+ weeks, with the brain updated from checked results.
  • Number gap: no metric of any kind on the site [site]. To own it we need one before/after number a buyer would sign.
  • Buyer gap: not shown. They say enterprise [yc] but sell nothing visible to anyone yet. We can not claim this gap until we see who their design partners are.
  • Price gap and speed gap: UNVERIFIED. Nothing public on price or time to first value. Do not use.

8. For the YC application

  • Proven today: Marton's logs show his system already runs agents that do and check work: about 5,600 jobs, each checked by a different AI, with about 160 failure records that each added an automatic check. Savant's public pages describe serving context to agents, with no claim that it runs or checks their work [site].
  • Draft answer to "how are you different from Savant?": DRAFT. "Savant stores a company's unwritten rules and hands them to agents. Our system already runs the agents, has a second AI check the work, and learns from what fails, every day, on our own business." Founders say it their own way.
  • Future claim (needs proof): "We learn the rules from checked results, not just from reading the company's tools." Needs one customer workflow run for 4+ weeks, with a rule-change log tied to checked runs and rework falling.
  • What we understand that they don't: a company brain gets better from the results of checked work, not from capture alone. needs proof on a customer; Marton's logs show it only on his own work (offer feedback lines fell from 44 to 18).
  • Copyability (M3): medium. Evidence: they already sit at the moment an agent decides [site], so feedback on whether the decision worked is a natural next feature. But running agents and measuring results is months of work and a different product from capture (inference). They are two people with no customer, so a quick pivot is also possible (inference). Not hard, so M3 can not score above 3 against Savant.
  • Why now (M4): None found. Do not use "agents are better now" or "everyone is adopting agents".
  • Do not claim: "nobody else builds a company brain" (Savant, Hyper and Memory Store are in the same batch, per the brief). "Our brain learns from checked results" as a customer fact (no customer yet). "We beat them on enterprise or SME" (their buyer is not shown).

9. Ideas worth taking

  1. "Nothing to migrate. No new place for people to live." A clear promise that lowers the fear of a rollout [site].
  2. Name the questions the brain answers in the buyer's words: refunds, pricing exceptions, "what an agent must never do alone" [site]. Concrete examples sell better than the word "memory".
  3. Say it is a private beta for design partners in the terms [site]. It sets honest expectations and makes the first customers feel chosen.

10. Unknowns

  • Who their design partners are, and how many. Look for a YC Launch post, a demo video on their YC profile, or a founder post (public web only).
  • Whether Kevin Yu is a co-founder (a directory says so; the YC page we read shows one founder).
  • Whether capture is passive reading, staff interviews, or both. A demo or the Guides page once filled would show it.
  • How agents call it (MCP, API) and what comes back.
  • Price, security proof (no certification named) and funding.
  • Whether the "AI-native enterprise" buyer holds, or whether they move to smaller firms.

Sources

Technical detail
Slug
savant
Lens hash
179096c479
Pass
deep
Stale
no