Textile.dev

Why we invested

INFERRED — NOT FROM MARK We hold no stated logic on this deal, so the following is our honest reconstruction of a plausible thesis, not a quote. Read it as "how this could fit Daxos," not "why Mark did it."

• CLAIMED THESIS Timing on a real anxiety. As people paste more into cloud AI, a backlash toward "nothing leaves your device" is a credible wedge — especially for documents users refuse to hand a tech company.
• VERIFIED Very cheap entry: ~$225K in at the earliest stage is a low-cost option on a possible category, fitting our small-check, high-optionality posture.
• INFERRED Warm, network-sourced dealflow: a UMiami-origin team overlaps our active Miami recruiting channel — proprietary and non-auction, which we like.
• VERIFIED Talent signal: one founder heading to MIT CSAIL, an EPIC competition win, USTAAR-backed — strong technical raw material for a hard problem (on-device inference plus retrieval).

What the company is

VERIFIED Textile is a consumer/prosumer desktop app: upload your documents, then semantic-search and ask questions about them, all processed locally with nothing sent to a cloud. It came out of the University of Miami "Bonsai" applied-CS group, is in beta, and is pre-revenue. VERIFIED Identity is resolved with high confidence: this is a new 2025/26 company, distinct from the unrelated legacy "Textile.io" IPFS project. UNVERIFIED Harmonic lists 95 headcount, almost certainly a name-collision error — a student seed company has a handful of people; discard that number.

What we probed in diligence

No data room and no saved Q&A — these are the questions this deal has to survive.

1. Moat. On-device document AI is increasingly commodity (Apple Intelligence, Microsoft Recall, local vector DBs). What is defensible beyond UX and a privacy promise?
2. Market reality. Is a "private personal document assistant" a paid consumer category, or a feature the OS platforms absorb for free? Willingness-to-pay is unproven.
3. Monetization. No pricing on the site; local-only removes the usual cloud subscription hook. One-time license, subscription, how do you retain and upsell?
4. Team durability. Student founders splitting across cities (one to MIT CSAIL) — full-time committed, or a strong side project?
5. Funding truth. Harmonic shows $225K/1 round; the web cites a $100K USTAAR award plus "venture-backed." Reconcile the real post-money and cap table before sizing.

Bull case

• Rides a durable, growing distrust of cloud AI with the clearest possible promise ("nothing leaves your device") on the highest-sensitivity documents.
• Very cheap entry (~$225K seed) on genuinely capable technical founders.
• Local-first is hard to do well — if the UX is great, it's a wedge into a trusted personal-data layer.

Bear case

• Platform-absorption risk: Apple/Microsoft/Google ship "search your files privately" for free, killing standalone willingness-to-pay.
• Thin moat — on-device retrieval is increasingly off-the-shelf; differentiation is UX, which is copyable.
• Student-founder, pre-revenue, split team; execution and full-time commitment unproven.

Key risks

• Commoditization by OS incumbents — the biggest one.
• Monetization undefined — local-only removes the usual cloud subscription hook.
• Consumer distribution / acquisition cost for a privacy tool with no obvious viral loop.
• Data reliability: local models getting tax, medical, or legal answers wrong carries real user-harm and trust risk.