perceptron

What it is and what it does

Perceptron is a Python SDK for physical AI—robotics, manufacturing, logistics, and security applications—that wraps access to grounded perception models (Perceptron Mk1 for cloud, Isaac family for edge). It provides a unified interface for detection with structured bounding boxes, image captioning, OCR, and visual Q&A, all with spatial grounding and in-context learning from annotated examples. The SDK abstracts provider selection (defaulting to Perceptron but supporting swappable backends), handles coordinate normalization and streaming responses, and includes a CLI for batch processing.

The package depends on rich, typer, Pillow, numpy, httpx, shellingham, colorama, and pydantic—a moderate but manageable set of dependencies. It targets modern Python (3.10+) and is designed for both cloud workloads (where capability matters) and edge deployment (where latency and footprint are critical). Configuration is flexible: environment variables, code-level context managers, or CLI commands. When API credentials are absent, the SDK returns compile-only payloads so you can inspect requests before sending them.

Use it for:

• Detect defects or safety violations in manufacturing images with few-shot learning from annotated examples.
• Build robotics pipelines that localize objects and obstacles with grounded spatial reasoning across video frames.
• Extract text and structured data from documents or schematics using OCR with custom prompts.
• Run real-time perception on edge devices (e.g., Isaac 0.2 2B) with low latency and minimal serving cost.
• Compose multimodal workflows combining detection, captioning, and Q&A using the typed DSL.
• Batch-process image directories with the CLI to generate JSON summaries of detections or captions.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

Perceptron is a Python SDK for building physical AI applications with multimodal perception tasks—detection, localization, OCR, and visual Q&A—routing requests to specialized models with structured outputs ready for robotics and edge deployment.

Yes, with conditions. Perceptron is actively maintained, has no known vulnerabilities, and low install friction. It is worth installing if you are building physical AI applications (robotics, manufacturing, security) and need a unified SDK for grounded perception. However, verify the license terms before production use—the metadata does not specify whether it is open-source, proprietary, or commercial. Also confirm that the model families and API access you need are included in your intended deployment model.

Install

perceptron on PyPI

pip install perceptron

uv add perceptron

poetry add perceptron

Installing perceptron

Before you install

Low friction: pure Python wheel with eight runtime dependencies (rich, typer, Pillow, numpy, httpx, shellingham, colorama, pydantic). Active maintenance since first release in September 2025, last updated May 2026. No known vulnerabilities.

License in practice

License status is unclear—no SPDX identifier or raw license text provided in metadata. Verify the actual license terms before deploying in production or redistributing.

Quickstart

pip install perceptron

from perceptron import detect, caption

result = detect("image.jpg", classes=["object"])
for box in result.points or []:
    print(f"{box.mention}: {box.top_left}")

desc = caption("scene.png", style="detailed")
print(desc.text)

Requires Python 3.10 or later. API credentials (PERCEPTRON_API_KEY) needed for actual inference; SDK returns compile-only payloads when missing.

Verify before relying

• Whether license is proprietary, open-source, or commercial—metadata does not specify
• Whether Perceptron Mk1 and Isaac model families are included or require separate subscription
• Actual latency and cost characteristics for edge models (Isaac 0.1, Isaac 0.2 2B)
• Whether optional extras (torch, dev) are freely available or require additional dependencies

Package facts

Evidence: perceptron-0.3.5-py3-none-any.whl

Tags

More Artificial Intelligence packages