Plumerai Documentation¶

Technical documentation for Plumerai Video Intelligence, a complete AI stack for consumer cameras, video doorbells, and commercial security cameras. For company information, visit our main website and for updates visit our blog.

Plumerai Video Intelligence¶

Plumerai Video Intelligence provides fast, accurate detection of people, vehicles, packages, and animals, and supports advanced features such as Familiar Face Identification and Vision Language Model (VLM) based capabilities that can run on-device, on-premise, in the cloud, or a hybrid combination of these. It has been deployed at scale and field-tested on millions of cameras across dozens of different SoCs, with models highly optimized for efficiency on everything from low-power edge hardware to cost-effective cloud compute, and integrated through a single unified API.

Components¶

Platforms, API, and deployment¶

Plumerai Video Intelligence runs on a wide variety of platforms and operating systems while using the same unified API across environments.

Because the library is built for your target platform, from microcontrollers to x86_64 servers, the same pipeline and API run unchanged whether a stage executes on the camera or in the cloud. Depending on the compute, bandwidth, cost, and privacy trade-offs, you can run the end-to-end pipeline at the edge or partition it across devices:

• At the edge. The full pipeline runs on-device, so image data never leaves the camera and no server connection is required.
• Hybrid edge and cloud. Lightweight components such as object detection and VLM video collection run on the camera, while heavier stages such as the VLM embedder run in the cloud. This cuts bandwidth and cloud compute cost while still offering the advanced VLM features.
• Fully in the cloud. For cameras or systems that cannot run AI on-device, the entire pipeline runs in the cloud on standard x86_64 servers (Intel Xeon or AMD Epyc with AVX-512), processing live streams or recorded video clips. Deployments typically run on AWS, Azure, or Google Cloud.

In multi-camera products, state such as Re-Identification identities can be shared between cameras through a cloud backend or a local hub.

API reference¶

The Video Intelligence API takes a video feed and returns objects, faces, and events. New integrations should start with the integration checklist, which flags the common mistakes that quietly degrade accuracy.

Language APIs: C++ · C · Python · Rust · Java

Examples and tests: Examples · Unit tests

Demos¶

We provide demo applications that run:

• on Arm Cortex-A and x86 CPUs
• on an ESP32-S3 microcontroller
• on a Realtek Ameba Pro2
• directly in your browser

Getting the software and support¶

The Plumerai Video Intelligence library is distributed as a release package built for your target platform. To request a package, or if you have a question that these pages do not answer, contact us at [email protected].

Plumerai Inference Engine¶

Separately from Video Intelligence, Plumerai offers the Plumerai Inference Engine: a deep learning model compiler and runtime for Arm Cortex-M and other microcontrollers that efficiently runs any quantized AI model. See the Inference Engine documentation, the blog post, or try it with your own model.