Introducing Qualcomm IMSDK 2.0 - a unified framework for AI and multimedia product development on Qualcomm Edge AI platforms

The new agentic skills in Qualcomm IMSDK 2.0 not only helps enable you to quickly assess the evaluation kit (EVK) platform best suited to your product, but also facilitates you with agentic/rapid prototyping. The SDK includes sample applications you can refine with the agentic coding skills.

Product development teams can then quickly move from assessment to deployment and scaling without having to rebuild. The SDK is trusted by development teams in companies such as Samsung, Amazon, and Bose.

Containerized microservices

Qualcomm IMSDK 2.0 extends beyond application development to provide a set of ready-to-use blueprints and containerized microservices. With the building blocks in the SDK, you can help optimize your own pipelines for performance, then deploy them as containerized microservices in three categories: inferencing, analytics, and platform.

Inferencing services support:

Vision AI	Generative AI	LLMs and VLMs
Audio AI	Text-to-image generation	Hardware-accelerated execution using Qualcomm AI technologies

For instance, inferencing microservices are compatible with the OpenAI API for features like chat completion. Features slated for subsequent release include Responses API and model lifecycle management.

Analytics services are strcutured to transform AI outputs into actionable insights in use cases such as:

People analytics	PPE compliance	Occupancy       monitoring
Vehicle analytics	Region monitoring	Heatmap generation

Platform services are developed to connect edge applications with enterprise and cloud systems through integrations with the following:

Kafka	MQTT	AWS IoT
Azure IoT	Confluent Cloud	Enterprise event-processing systems

Qualcomm IMSDK 2.0 microservices are building blocks you can use to compose and deploy applications. You can combine inference services and analytics services, add APIs, messaging systems, and cloud connectors to create complete enterprise solutions.
You get rapid innovation without starting from scratch, while preserving scalability and maintainability.

Solutions include:

Industrial safety monitoring: Person detection, PPE compliance analytics, alerting APIs

Restricted zone monitoring: Region analytics, occupancy alerts, person detection

Vehicle analytics: Vehicle detection, parking analytics, cloud event forwarding

Edge conversational AI: LLM/VLM inference, camera context, application APIs

Multilingual audio assistants: Speech recognition, translation, text-to-speech

Enterprise edge analytics: AI inference, analytics services, Kafka or MQTT integration

AI-powered products that Qualcomm IMSDK 2.0 enables

Using Qualcomm IMSDK 2.0, you can build applications around cameras, video streams, audio processing, AI inference, advanced analytics, and cloud connectivity within a single architecture for products like these:

• Smart cameras
• IP and web cameras
• Industrial vision systems
• Robotics platforms
• Autonomous drones
• Edge AI boxes
• Worker safety solutions
• Smart retail applications
• Intelligent surveillance systems
• Multimodal AI assistants
• Edge analytics platforms 

What’s new in Qualcomm IMSDK 2.0?

Readable Pipeline APIs and app builders for Python and C++

In a single public GitHub repository we’ve consolidated the core Qualcomm IMSDK:

GStreamer plugins – The foundation of Qualcomm IMSDK is a GStreamer-based multimedia and AI framework with Qualcomm hardware-accelerated plugins, zero-copy data paths, and flexible support for multi-stream and multi-model pipelines.

App builders – For development teams without deep GStreamer expertise, Qualcomm IMSDK 2.0 extends that foundation with Python and C++ app builders for working directly in those languages (see below).

Developer tools – The repository also includes utilities, command-line helpers, code, examples, and tools to build applications.

Documentation as code

Qualcomm IMSDK 2.0 treats documentation as code linked directly with corresponding code repositories, so samples, API guidance, pipeline instructions, and implementation assets are aligned.

You’ll find the documentation oriented toward how-to and blueprints, so you can understand how to combine the available components to build complete application workflows.

Plus, the documentation is delivered through both a dedicated public GitHub repository and the external Qualcomm IMSDK site.

Coding agent skills

Qualcomm IMSDK 2.0 provides its capabilities as reusable coding agent skills through GitHub. The coding agent lets you build end-to-end AI applications in an agentic, code-first workflow that includes the supporting tasks required to compile, run, debug, and deploy your applications.

You can install or reference those skills from your preferred coding agent, such as Claude Code. You can then use natural-language prompts and code-based workflows to create Qualcomm IMSDK applications, configure pipelines, develop plugins, on-board AI models, post-process, and troubleshoot integration issues. The skills are intended to help the coding agent understand Qualcomm IMSDK APIs, plugins, microservices, and deployment requirements to generate and connect the components.

Choose your inference runtimes, choose your models

Multiple inference runtimes

Qualcomm IMSDK 2.0 lets you change the model and/or inference route without having to recode the camera, multimedia, analytics, and deployment pipeline around them.

As depicted below, the Runtime Inference Plugin supports runtimes like QAIRT, ONNX Runtime, and TFLite. You can choose the runtime that best matches your workflow and product, while keeping the surrounding multimedia and application pipeline (top row of Figure 5) consistent. 

You can also map the runtimes to CPU, GPU, or NPU (bottom row of Figure 5). That gives you flexibility among runtimes like the Qualcomm AI Runtime SDK (QAIRT), ONNX and TFLite for vision and GenAI models.

Besides the AI stage, Qualcomm IMSDK supports the surrounding steps: model input preparation, image transformation, batching, inference, tensor post-processing, metadata composition, overlays, analytics, and actions.

Multiple model sources

With Qualcomm IMSDK 2.0 you can select your model sourcing and model development path.

• Qualcomm AI Hub helps you select a model optimized for the Qualcomm platform you’re targeting. You’ll find compilation options, runtime support, and performance and accuracy results.
  
  
• The Qualcomm Technologies page at Hugging Face includes Qualcomm-published models and can connect you to the broader open-source model ecosystem.
  
  
• For industrial and domain-specific applications, Edge Impulse lets you collect and label your own data, train models, and optimize those models for edge deployment.
  
  
• Finally, you can Bring Your Own proprietary or externally sourced Model (BYOM). Qualcomm AI Hub recipes and applicable Qualcomm workflows (including conversion, quantization or precision selection, compilation, execution testing, and accuracy validation) can help prepare models for your selected runtime and target.

Pipeline APIs make AI and multimedia code easier to read

As described above, the Pipeline API for Python and C++ lets you describe an application in terms of the pipeline you’re building, all the way from creating the pipeline to managing it. A readable pipeline is easier to review, debug, modify, explain to another engineer, and extend from a prototype into a product.

Python for faster experimentation

Choose the Qualcomm IMSDK Python app builder to build, run, and manage multimedia and AI pipelines in Python on Qualcomm platforms. It provides a high-level abstraction layer over GStreamer that lets you focus on application logic rather than on framework.

C++ for production integration

The C++ Pipeline API brings the same structured model to applications that require native performance, tighter product integration, and explicit control over lifecycle and execution. You can use the app builder without the need to implement every GStreamer detail directly.

The following table compares the Qualcomm IMSDK 2.0 Pipeline API against the traditional approach to integration:

Concern	Traditional approach	Qualcomm IMSDK 2.0 Pipeline API
Expressing the flow	Many low-level objects, properties, callbacks, and framework-specific setup steps.	Pipeline structure is visible: create, add, connect, configure, run.
Learning curve	Developers may need deep knowledge of the multimedia framework before building an application.	Python and C++ app builders provide a higher-level starting point.
Readability	The application intent can be spread across initialization and helper code.	Data flow and component relationships are easier to see in one place.
Changing a model	Inference changes can require changes across model loading, buffers, pre-processing, and output handling.	Inference components can be swapped while the surrounding pipeline remains recognizable.
Debugging	Failure may be distributed across framework calls and runtime configuration.	A structured pipeline makes it easier to isolate the stage that needs attention.
Moving to deployment	Build, package, and target-device steps can become a separate integration effort.	The SDK ecosystem connects app building with deployment, Docker, coding agent, and device-in-loop workflows.

What’s in the Pipeline APIs for you?

The new API is designed to reduce accidental complexity so you can focus on what the application should do while the framework provides a consistent way to compose the pipeline.

• No GStreamer expertise required – Qualcomm IMSDK 2.0 manages all underlying GStreamer mechanics internally, with no compile-time dependency on GStreamer headers.
  
  
• Fast pipeline development – You can write a functional camera-to-display pipeline in just a few lines of Python or C++.
  
  
• AI inference support – Qualcomm IMSDK 2.0 includes built-in binaries for TFLite, QNN, SNPE, and ONNX models, with support for custom pre-/post-processing.
  
  
• Two pipeline definition styles – Define pipelines programmatically using the Python or C++ API (add() / link() calls), or declaratively using a YAML configuration file passed to the Pipeline constructor.
  
  
• Broad multimedia flexibility – Qualcomm IMSDK 2.0 supports video capture, signal processing, machine learning inference, overlay rendering, and display output.

Next steps

If you develop, compose, and deploy AI and multimedia products on Qualcomm edge platforms, Qualcomm IMSDK 2.0 provides building blocks that help enable you to focus on building applications instead of integrations.

See how other developers have used Qualcomm IMSDK in defect detection, factory digital twin, and worker safety. Then explore the documentation.

When you’re ready, get Qualcomm IMSDK 2.0 from the Software Center.

Qualcomm branded products are products of Qualcomm Technologies, Inc. and/or its subsidiaries.