Perplexity unveils hybrid agentic inference: What it is and why it matters

AI company Perplexity has announced a new feature called "hybrid agentic inference" for its Personal Computer platform, aiming to solve one of the biggest challenges facing artificial intelligence today: deciding what should run on your device and what should run in the cloud.
The company says the new system will automatically split AI tasks between local models running on a user's computer and larger, more powerful models operating in remote data centres. The goal is to improve privacy, reduce costs and deliver better performance without requiring users to make technical decisions themselves. The announcement comes as technology companies increasingly look for ways to combine the speed and privacy of on-device AI with the power of cloud-based systems.

What is hybrid agentic inference?

Hybrid agentic inference is a system that allows AI workloads to be shared between a user's device and cloud-based AI models. Instead of sending every request to a remote server, Perplexity's system first evaluates the task and decides where different parts should be processed. For example, if a user asks an AI assistant to analyse financial records, personal documents or health information, the sensitive data can remain on the device. More complex reasoning tasks can then be sent to larger cloud-based AI models when additional computing power is needed. According to Perplexity, the entire process happens automatically without requiring users to choose between local and cloud processing.

What is the 'orchestration problem'?

At the centre of Perplexity's announcement is a concept known as the orchestration problem.

As AI systems become more advanced, companies now have access to multiple models with different strengths. Some models are small and efficient enough to run on personal devices, while others require large-scale cloud infrastructure.

The challenge is deciding:

Which model should handle a task
Where the task should be processed
How data should move between systems
When cloud resources are actually needed

Perplexity says hybrid agentic inference is designed to solve this problem automatically.

In simple terms, the system acts like an intelligent traffic controller, deciding which AI model is best suited for each part of a request.

Why is Perplexity focusing on local AI?

The company believes that modern computers are becoming powerful enough to handle a much larger share of AI workloads locally. Over the past few years, improvements in processors, graphics chips and dedicated AI hardware have made it possible to run increasingly capable AI models directly on laptops and desktops.

Perplexity argues that local AI offers several advantages:

Better privacy
Faster response times
Lower cloud computing costs
Reduced dependence on remote servers
Greater user control over personal data

The company summed up its philosophy by saying that people would rather own "a data centre in their laptop" than depend entirely on infrastructure controlled by technology companies.

How does the system protect sensitive data?

Privacy is one of the main reasons behind the hybrid approach. According to Perplexity, a compact AI model running on the user's device first reviews requests and determines whether sensitive information should remain local.

This could include:

Financial records
Medical information
Personal documents
Private business data

Only tasks that require additional reasoning or processing power are sent to larger cloud models. The company believes this approach can reduce the amount of personal information transmitted to remote servers while still giving users access to advanced AI capabilities.

Partnerships with Intel and Nvidia

Perplexity introduced the technology alongside Intel and says the platform has been designed to work across multiple hardware ecosystems. The company also highlighted support for Nvidia's RTX Spark platform, one of the latest AI-focused computing architectures aimed at running advanced AI models locally. Importantly, Perplexity says its orchestration layer is model-agnostic. This means it is not tied to a specific AI model or hardware platform and can potentially operate across a range of processors, chips and AI environments.

How does it compare with Apple, Google and Microsoft?

Perplexity is not the only company pursuing a hybrid AI strategy. Several major technology firms have already introduced similar approaches:

Apple
Apple Intelligence combines on-device AI processing with Private Cloud Compute, allowing users to access larger AI models while keeping sensitive information protected.

Google
Google offers Gemini Nano for local AI processing on supported devices while relying on larger Gemini models in the cloud for more advanced tasks.

Microsoft
Microsoft has expanded local AI capabilities through its growing family of AI models and AI-focused Windows hardware.

Perplexity argues that its key difference is automation. Rather than asking users or developers to decide where workloads should run, the system makes those decisions automatically as part of a single workflow.

Why this matters for the future of AI

The announcement highlights a growing shift in the AI industry. For years, the dominant model involved sending nearly all AI requests to large cloud data centres. However, the cost of operating advanced AI systems continues to rise, while concerns around privacy and data ownership are becoming more important. Hybrid AI is increasingly viewed as a practical solution because it combines the strengths of both approaches.

Industry analysts expect future AI assistants to operate across multiple layers, using local processing whenever possible and cloud resources only when necessary.

When will it be available?

Perplexity says support for local inference on its Personal Computer platform is expected to begin rolling out in July. The company has not yet disclosed full hardware requirements or confirmed which devices will receive support first. However, the announcement suggests that Perplexity is positioning itself alongside Apple, Google and Microsoft in the growing race to make AI more personal, private and efficient. As AI becomes a larger part of everyday computing, systems capable of intelligently balancing local and cloud processing could play an increasingly important role in how users interact with technology.