Clairvoyance integrates GenieX for local Agentic AI tasks on Snapdragon X Series
TL;DR:
- To run Agentic AI efficiently on a local laptop, you have to match the difficulty of the task to the model size and the compute core.
- Clairvoyance abstracts the details of running local AI tasks efficiently through an agentic workplace that aligns tasks, models and compute cores so that users don’t have to.
- Through GenieX, an on-device GenAI inference runtime built for Qualcomm platforms, Clairvoyance is now integrated with Hexagon NPU. Enterprise users enjoy native-built capabilities for running a wide range of AI tasks locally on CPU, GPU, and NPU.
Are you looking for the best way to run your AI models locally? That goal is now within easy reach as more OEMs build laptops around Snapdragon X Series, with neural processing units (NPUs) designed to run AI workloads.
That’s why Clairvoyance AI used GenieX to run small and large language models on Hexagon NPU with their agentic workplace. Clairvoyance lets you launch, orchestrate, and monitor a staff of AI assistants inside designated areas of your local computer.
Natively built for Windows on Snapdragon, Clairvoyance can now access up to 228 GB/s of memory bandwidth and 80+ tera operations per second (TOPS) on the Snapdragon X2 Elite Extreme.
Read on to learn how the team used GenieX to perform the integration. You’ll also get insight into the best ways to run workloads locally on NPU as part of your agentic AI deployment strategy.
Integrating GenieX
GenieX is an on-device GenAI inference runtime. It’s designed for Snapdragon platforms like Snapdragon X/X2 Elite (computers), Snapdragon 8 Elite (mobile devices), and Qualcomm Dragonwing (IoT). You can use GenieX to run frontier language and vision-language models locally on Hexagon NPU, Qualcomm Adreno GPU, or CPU with just a few lines of code.
“Clairvoyance runs agents that use tools and coding models,” says Brad Wardell, founder and CEO of Clairvoyance AI (the company). “GenieX is open source, so we customized it to our needs. And the way it’s set up, we had GenieX running workloads on the Qualcomm Hexagon NPU on Snapradgon X2 Elite within a day.”
You have the choice of five options for installing GenieX:
- Command-line interface (CLI)
- Local server
- Python SDK
- Linux (Docker)
- Android SDK
You specify where you’ll run GenieX (Windows ARM64) and which runtime you need. You can deploy AI models from Qualcomm AI Hub or Hugging Face, or you can Bring Your Own Model (BYOM).
“The integration was seamless,” says Brad Wardell. “We used AI to handle almost all of it. “We chose the CLI install from the GenieX documentation that Qualcomm Technologies provides. The quick start page had everything we needed to get GenieX running. I’d call it plug-and-play.”
Once GenieX was running on Snapdragon X2 Elite, it opened a port and created a local server for NPU inferencing. Clairvoyance team immediately started using GenieX to run SLMs/LLMs on the device and for agentic tasks such as generating and running detailed graphical displays like dashboards, artifacts, and canvases in HTML.
Snapdragon_GenieX_Demo
Sep 23, 2026 | 12:07
Integration results
The integration with Hexagon NPU on Snapdragon X Series paved the way to the three biggest advantages of running AI locally:
- Privacy – You keep the data out of the cloud.
- Latency – You eliminate the round trip needed when compute is in the cloud.
- Cost control – Local models mean local tokens. That means you’re not paying for someone else’s system to generate them for you.
But not all AI workloads make the same demands on hardware resources. More to the point, not all users think (or care) about workloads and resources, nor should they be expected to.
Clairvoyance is positioned as the easiest way for users to run local AI because it does that thinking for them. It looks at the hardware and the task requirements, then it picks the best combination of model, schedule, and hardware resources for the agentic task. Like any AI developer responsible for smart AI usage, Clairvoyance—assisted by Snapdragon X2 Elite and the Hexagon NPU—drills into two important questions.
Question 1: What is the task?
Clairvoyance AI has identified several levels of tasks, based on the resources required to deliver an acceptable outcome:
- Questions, quick answers, brainstorming, conversing. "Rewrite this sentence."
- Looking things up. "Find my Q3 report and show it to me."
- Making and editing things. Creating and updating CSVs, JSON, documents, and exhibits.
- Reading hard material, writing about it in plain language.
- Working across services with follow-through. "Watch my email for anything from XYZ Corp and notify me."
- Writing and running code. Source edits, builds, version-control changes, code review.
- Deciding what work should even happen. Planning and decomposition.
Those task levels, in turn, are aligned with the number of parameters in the model running locally:
- Conversing and looking things up – 3 Billion (3B)
- Making and editing – 9B
- Reading hard material – 13B
- Working across services – 27B
Understanding those levels leads to the next question.
Question 2: Is the hardware up to the task?
Consider the example of chat and conversation with a model, which is a common level-1 use case. For Clairvoyance, Snapdragon X Series optimizes the two steps in that task.
“Step one is the prefill,” says Brad Wardell, “when the model reads your prompt and the entire conversation. That’s a good use of Hexagon NPU. TOPS matters because conversations with AI get better with more context; without NPU or high TOPS, the longer the conversation goes, the slower the conversation gets.
So, with 80+ TOPS on the Qualcomm Hexagon NPU, Snapdragon X2 Elitecan prefill faster than any standard laptop, and the result is a great chat experience for users.
Then, step two is decoding, or writing the answer to the prompt, and the Qualcomm Hexagon NPU is well suited to that kind of work. Best of all, GenieX is smart in how it manages the workload distribution between CPU and NPU.”
According to Clairvoyance AI, depending on the hardware configuration, AI developers can run levels 1 through 3 entirely locally with acceptable results. In fact, Clairvoyance running on Hexagon NPU delivers excellent results all the way up to level 5.
Multiplayer AI on Snapdragon X Series
Clairvoyance provides multiplayer AI on Snapdragon X series devices which enables teams and AI agents to work together in a shared workspace, combining human creativity with persistent, on-device AI assistance powered by the Hexagon NPU.
The advantages for users include the following:
- Shared workspace for people and AI agents
- Multiple specialized AI staff members collaborating on tasks
- Persistent memory across projects and sessions
- On-device AI acceleration for speed and privacy
- Intelligent orchestration between local and cloud AI
- Faster decision-making, content creation, and execution
- Reduced cloud dependency with always-available AI assistance
Next steps
For running agentic AI tasks on laptops powered by Snapdragon X Series, the guidelines in Clairvoyance AI’s Local AI Primer apply:
- 3B models on NPU for chat (levels 1 and 2)
- 9B models on NPU for light tool use (levels 3 and 4)
- 27B models on NPU for high-compute work (levels 5 and above)
Visit the GenieX GitHub repo and documentation to learn how you can integrate your applications with Hexagon NPU.
See our Qualcomm AI Hub community on Slack for more information. And browse through models of laptops with Snapdragon X/X2 Elite.
To get started with AI app development on Snapdragon X Series, visit the Windows on Snapdragon documentation.
