Investors

The layer every machine will need.

Every device that listens or looks will need one thing: a way to decide, in the moment, what it may keep, and to prove it. We are building that layer. The patents were filed first, the engineer who built the incumbent's core technology is building ours, and the first regulated buyers are already in the pipeline. If you back category-defining infrastructure early, we would like to hear from you now.

Introduce yourself: invest@cull.ai

One line about your fund and why this thesis fits. We reply fast.

The pattern, proven twice

ElevenLabs owned the primitive

One company owned the hard voice problem; every product bought it as infrastructure rather than building it.

Anthropic made safety the company

Governance was not bolted onto someone else's model. It was the thesis, the moat, and the reason to fund it.

Cull is the layer between them

Policy-based governance of everything machines hear, decided before anything is written, and patented.

The strategy is talent aggregation: the world's best voice, speech, and diarization engineers, in one company, building the method the world verifies and adopts.

Standing on open source where it is strong. Customizing the stack wherever trust must be earned: streaming attribution with no lookahead, judgment inside the buffer window, absence you can prove.

Staged capital, like the labs we admire
NOW · SEED
Prove it in production. The consent-gated meeting agent live with paying design partners in financial services and legal, and the first patent moving to issuance. PatentVC has committed capital as needed to reach each stage: outside capital is for acceleration, not survival.
12–18 MONTHS · SERIES A
Build what makes it defensible for a decade. Our own governance-tuned speech models, the on-device and in-tenant stack that runs with the network cable pulled, and the hardware attestation path.
24–36 MONTHS · PLATFORM
License it into everything. Per-unit economics on wearables, vehicles, and humanoid fleets: the margin of infrastructure on someone else's hardware.
What a partner sees in diligence
01 · The team's record

The engineer who built the incumbent's speech AI, named on 12 of its 19 public patent filings. A Stanford research scientist who has shipped regulated AI three times. A studio founder whose patents exited to Google and license to Nextdoor today.

02 · The filings

Six accelerated Track One applications spanning the governance stack, drafted apparatus-generic so the claims travel from meetings to robots, with the parent on file before the litigation reshaped the category.

03 · The pipeline

A beachhead where the buyer is the general counsel and the alternative is unlawful: financial services first, legal second, healthcare behind them, with the compliance report as the leave-behind.

04 · The demo

Kill the process mid-meeting: zero recoverable audio or transcript of any non-consented speaker. The patent choke point, live, in ten seconds.

What we think about at night, and tell you about in the first meeting
Clean-room discipline is existential. Our CTO's history is the moat and the exposure at once, so it is governed like a control, not a guideline.
Enterprise sales craft is a known gap. None of the founders is a quota-carrying closer; we budget the hire instead of discovering the gap at quota time.
The humanoid timeline may slip. Robotics is the option, not the plan; the regulated-enterprise business is priced to stand alone.
Transient processing is a strong legal position, not a settled one. Edge and in-tenant execution exist to keep it that way.

Partners who find that list reassuring rather than alarming are the partners we want.

Introduce yourself: invest@cull.ai
Next: Home →