Plumerai

Market position

The company describes itself as a pioneer in on-device AI for cameras and the market leader in licensable Tiny AI for home security cameras. It cites customer-run benchmarks showing its AI Video Search as more accurate than cloud-only alternatives such as Amazon Nova and Google Gemini, with cloud costs up to 135x lower. These claims are company-sourced.

Vertical integration across model architectures, inference engine, proprietary datasets and Vision Language Models into a single end-to-end optimized pipeline; the ability to run accurate models on sub-$1 chips with 1MB of memory and on battery power; and the hybrid architecture in which on-device Tiny AI gates when cloud Vision LLMs are invoked, cutting cloud inference cost while retaining rich scene understanding.

Technology

Plumerai builds its neural networks from scratch, using customized model architectures and training strategies for embedded devices grounded in research on model quantization. The Plumerai Inference Engine runs models efficiently on CPUs, optimized for Arm Cortex-M, Arm Cortex-A and RISC-V, with additional support for x86 (SSE2 through AVX512), Arm Ethos-U55/U65, ESP32-S3, ARC EM, ARM7/ARM9 and various NPUs, across Linux, Windows, macOS, RTOS such as FreeRTOS, and bare metal. Published latencies include 18.1 ms for people detection on a Cortex-A72 (Broadcom BCM2711) and 12.0 ms on Arm Ethos-U55, and 26 ms to 250 ms for face identification across Ethos-U55, Renesas RZ/V2L and Renesas RA8. The company collects, labels and maintains its own dataset of over 100 million images and videos, using field-based "data actors" and a pipeline that surfaces failure cases. Its Multi-Camera ReID system uses an embedding network over person crops with tracking, enrollment and gallery matching stages, built on top of its people detection and advanced motion detection components.

Go-to-market

Sells through camera OEMs, smart home and security brands, and works closely with semiconductor companies whose microcontrollers and SOCs host its models; the website offers live demos, documentation and evaluation request forms, and the company publishes technical blog posts and benchmark comparisons. A key reference customer is Chamberlain Group (myQ, LiftMaster).

Camera makers and semiconductor companies, smart home camera and video doorbell brands, security camera and video analytics vendors, video conferencing and webcam makers, and builders of intelligent people-detection sensors; end markets cited include home security, enterprise security, retail, offices, warehouses and elderly care.

Geography

Headquartered in London with an office in Amsterdam and colleagues in the US; software is offered with compliance for US, EU and UK privacy regimes.

History

One source states Plumerai was co-founded in 2017 by Roeland Nusselder and Taras Iakymchuk. Total funding exceeded $17M as of the September 2025 Series A, which followed earlier rounds from investors described only as "existing investors." The company's early positioning centred on Tiny AI for edge cameras; in September 2025 it launched its first Vision LLM-powered features (AI Video Search and AI Captions) and broadened from home security into enterprise security and retail. A May 2026 engineering post describes shipping Multi-Camera Re-Identification, and a June 2026 post describes a company offsite in France attended by US, London and Amsterdam staff.

Risks & controversies

Performance, accuracy and cost claims (including the 135x cloud-cost advantage and "world's most accurate people detection") are company-published or based on customer benchmarks not independently verifiable from the sources. The product involves face identification and person re-identification, which are subject to biometric and privacy regulation; the company states compliance with GDPR, CCPA, BIPA and the EU AI Act. No controversies are reported in the sources.