VOwhalesYour platform for on-device automated, intelligent, and proactive assistance and agents that remember and seamlessly embed into your workflow. Leveraging a ExecuTorch's ability for efficient local model inference and combining it with heuristic AI systems and always-on knowledge ingestion, never lose anything again. Composed of the WhaleBrain to autonomously ingest and maintain knowledge, along with WhaleSense to proactively assist and embed itself into your system, and the WhaleRuntime that manages models exported and deployed in ExecuTorch, the Whale system provide the platform for future knowledge systems and seamless agentic experiences.
QualcommExecuTorchPython+2
Three AI tools that run entirely on a Snapdragon X Elite Copilot+ PC: no cloud, private by default. Each one solves a real-world problem. 1) NIGHTWATCH: night safety for memory care In 2019, Mildred Hernandez, 100, with Alzheimer's, walked out of her care home at night and died of hypothermia; a jury later awarded her family $110 million (according to the lawsuit). At night, one worker may cover about 40 residents, and door alarms get ignored. NightWatch runs on one laptop. An on-device person detector (SSDlite-MobileNetV3 via ExecuTorch) sees a resident cross their doorway at night; a second small model learned what a normal night looks like, so bathroom trips stay quiet. The staff phone gets an alert with what usually calms that resident. If nobody responds, it keeps escalating: urgent, then emergency, then family and facility owner at once (911 stays a human decision). It works with the internet off, records no video (the dashboard only sees positions), cameras can be off or night-only, and staff file the care record in two taps in the care team's own format. Same engine, wider use: hospital fall risk at night, group homes, aging parents living alone, privacy-first baby monitors. 2) INFERENCE RALLY: choose the right AI for robots Robots, carts and drones decide at every turn, and a slow decision is a crash. Inference Rally races real models live on the Snapdragon NPU through ExecuTorch + Qualcomm QNN (Gemma-4-E2B vs Qwen3-1.7B): at every junction the model picks a direction and the car waits while it thinks, so faster on-device inference visibly wins, with time-to-first-token, tokens/second and wrong turns measured. UNFAIR mode adds a rogue model that cheats (hedge jumps, drone scouting) when Laya, a decision model also on the NPU, says it will get away with it, and every cheat is counted. Use: picking model + chip before deployment, and testing how autonomous agents behave under pressure. 3) ALIBI BREAKER: check whether a story is physically possible A local model plays the suspect or claimant; a second, fast local model pulls every place and time out of each answer and pins it on a real San Francisco street map (OpenStreetMap). Plain code, not AI, checks each claim against walking times and evidence (e.g. "never left dinner", but a card receipt across the neighborhood) and flags what can't be true. The AI organizes; the math proves it; a human decides. Use: insurance claims investigation, internal and security investigations, checking a source's timeline. On-device because interviews and evidence are confidential. Honest notes: demo residents and cases are synthetic; NightWatch's camera walk-out can run in a labelled simulated mode; NightWatch currently runs its models on the CPU through ExecuTorch (NPU export path included); Inference Rally runs on the NPU.
QualcommExecuTorchQualcomm QNN / QAIRT SDK+11
CSecondHands"Beyond Weight. Measure What Matters" MotionVitals AI turns a camera and a Snapdragon device into a consistent, private measurement of trusted clinical mobility tests. Our initial target is the detection of muscle loss, a serious side effect of the GLP-1 drugs (Ozempic etc.), which more than 10% of the US adult population is on. So it's a problem than affects millions of people directly, and millions more indirectly. A pose model and a movement-phase model run on the Qualcomm Hexagon NPU through ExecuTorch; a rules engine turns body points into repetitions, timings and quality flags; nothing leaves the device and no video is stored. SecondHands is an affordable, adaptable AI-powered system that enables people with limited mobility to interact with everyday objects using whatever voluntary movements they can still perform. For someone living with paralysis, ALS, or another mobility-limiting condition, a small head movement, facial gesture, or hand motion could become a way to switch on a lamp, open a curtain, operate an appliance, or request assistance.
QualcommSnapdragon
An open-source chest X-ray reader that shows its work. A vision language model reads the film, reports the regions it looked at in order, and drafts a radiology report. The UI shows that reading path live as a red cloud moving over the image, then gives you an editable Exam / Findings / Impression, a one-click differential diagnosis and management panel, and a two-button rating that feeds a per-model benchmark. Works with any OpenAI-compatible endpoint (OpenAI, OpenRouter, vLLM, Ollama), the Anthropic API, or Amazon Bedrock. Ships with a mock engine so it runs with no API key, and a fully on-device local engine built on ExecuTorch (below). Reading Research prototype. Not for clinical use. The radiologist decides. OpenRead is for exploring how vision language models read radiographs. It is not a medical device, it has not been validated, and nothing it produces should be used to diagnose or treat a patient. Hackathon plan: the five-task pipeline, the Snapdragon ladder, the Astra-reference benchmark and dataset notes are in docs/PLAN.md. Run it fully on-device (ExecuTorch Hackathon) ENGINE=local replaces the single vision language model with a pipeline of small specialist models that run on the laptop with no network: a chest X-ray classifier exported to ExecuTorch, a reading path built from its activations, and a small local LLM (or a template) that writes the report. Built for the ExecuTorch Hackathon compute track (Dell Latitude 7455, Snapdragon X Elite, Windows on ARM): QNN/HTP for the NPU, XNNPACK on the CPU as the fallback. image | v [1] findings model ......... py/cxr_worker.py (Python, spawned once, JSON lines) TorchXRayVision DenseNet-121 "densenet121-res224-all" as an ExecuTorch .pte -> 18 pathology probabilities + a CAM box per pathology (no .pte? falls back to PyTorch on CPU and says so) | v [2] reading path ........... server/lib/findings.js (JavaScript, deterministic) lungs overview -> each positive finding with its CAM box, zoom: true -> heart and mediastinum -> pleura and costophrenic angles -> bones and soft tissues streamed as path events with a small delay, so the cloud animates | v [3] report writer .......... LOCAL_LLM= executorch ExecuTorch Llama runner (llama_main) with a 4-bit Llama 3.2 3B or Qwen3 4B openai-compatible any local server (Ollama, LM Studio, vLLM) for testing on a Mac template no LLM: a deterministic report straight from the probabilities -> token events while it streams, then report, then done The same step answers "Show me differential diagnosis" (rule-based with template). Three model roles, three files you can swap independently: role model where swap for findings TorchXRayVision DenseNet-121, 18 outputs, 224 px models/cxr-densenet121.pte via scripts/export_cxr.py any classifier that returns probs + feature maps (edit py/cxr_model.py) reading path none, rules over the findings server/lib/findings.js a segmentation model for true lung fields report writer Llama 3.2 3B Instruct or Qwen3 4B, 4-bit models/.pte via scripts/export_llm.md any instruct model the runner loads Quickstart on the Dell (details in scripts/setup-snapdragon.md) 0. the pipeline with no models at all (canned findings, templated report): proves the plumbing ENGINE=local CXR_WORKER=py/cxr_worker_mock.py LOCAL_LLM=template npm run dev 1. real classifier on CPU through PyTorch (downloads 30 MB of weights once) python -m venv .venv && . .venv/bin/activate # .venv\Scripts\activate on Windows pip install -r py/requirements.txt ENGINE=local LOCAL_LLM=template npm run dev # log says: falling back to PyTorch 2. export to ExecuTorch and run the .pte (XNNPACK; add --backend qnn for the NPU, see the docs) python scripts/export_cxr.py # writes models/cxr-densenet121.pte + verification run ENGINE=local LOCAL_LLM=template npm run dev # log says: findings worker ready, backend=executorch 3. a local LLM writes the report a) tonight on a Mac: ollama pull llama3.2:3b, then ENGINE=local LOCAL_LLM=openai-compatible LOCAL_LLM_URL= LOCAL_LLM_MODEL=llama3.2:3b npm run dev b) on the device with ExecuTorch: scripts/export_llm.md, then LOCAL_LLM=executorch LLM_RUNNER=... LLM_MODEL=... LLM_TOKENIZER=... 4. a labelled sample set and the agreement check python scripts/fetch_samples.py --n 40 # NIH ChestX-ray14 sample via Kaggle (see samples/README.md) ENGINE=local npm run bench # per-label agreement of the classifier's top finding All variables are documented in .env.example under "local". npm test runs the local engine through py/cxr_worker_mock.py, so the test suite needs Python 3 but no PyTorch. Honesty notes The classifier is a research model (TorchXRayVision, trained on public datasets with NLP-mined labels). Its probabilities are not diagnoses; they are the model's calibrated outputs on its own operating point, and they are quoted as percentages so nobody mistakes them for certainty. The reading path shows where the classifier's activation was strongest for each finding (class activation map, thresholded at 60% of its max, largest region). That is not a lesion outline and not eye tracking. The anatomic order around those boxes is fixed by us, not chosen by a model. The report writer has not seen the image. It rewrites the classifier's findings. Negatives are phrased "not flagged by the classifier", never "normal". Nothing here is a medical device. The radiologist decides. 5-minute quickstart git clone openread && cd openread cp .env.example .env # ENGINE=mock works with no key npm install npm run dev # API on :8787, web on :5173 Open click Try the sample image, then Analyze. You will see the mock engine stream a six-region reading path and a report. To use a real model, edit .env: OpenAI ENGINE=openai-compatible MODEL=gpt-4o OPENAI_API_KEY=sk-...
Qualcomm
DTXSplitLM rebuilds LLM inference on a Snapdragon X laptop around what each chip is good at, and pushes each phase toward its hardware ceiling. 1. Phase-split engine: long prompts are prefilled on the Hexagon NPU (about 4x the CPU's prefill speed), tokens are decoded on the Oryon CPU (which reads memory fastest), and the model state moves between them through shared memory in milliseconds. For big models the CPU decodes straight from the NPU's copy of the weights, so a 24B MoE (LFM2.5-24B-A2B) runs split in 14.7 GB instead of 27 GB. Concurrent requests overlap: the NPU prefills while the CPU decodes a batch. OpenAI-compatible server and live dashboard. 2. Custom kernels (llama.cpp fork): fused NPU element-wise kernels (+26% Qwen3.5-2B NPU prefill vs stock kernels), a streamed, pipelined NPU MoE expert matmul and causal block skipping in NPU flash attention (LFM2.5 prefill +21% vs stock, then +9-19% more), CPU kernels that read NPU-tiled weights in place, decode-overhead cuts, and MTP speculative decoding (Qwen3.8-27B 1.7x). Results on a Dell Latitude 7455 (X Elite X1E80100, 32 GB): - Qwen3.5-2B, 4K-token request: 2x faster than llama.cpp's default device choice, 1.4x NPU-only, 2.5x CPU-only (same build). - Demo: first token in 0.98 s vs 3.84 s CPU-only on a 1.6K-token document, same decode speed (48 t/s). - NPU prefill reaches 39% of the 11.48 TFLOPS FP16 peak (2B); 1.2x Qualcomm's own GenieX runtime on the same model file. Every speedup is a same-session A/B with identical text or perplexity; methods and limitations are in the repo.
Qualcommllama.cpp / ggml (fork with custom Hexagon NPU and Arm CPU kernels)Hexagon SDK 6.6+3
YktisBlindShare keeps a screen share safe. It captures a window on this PC, reads the text with Windows OCR, and uses an AI model to cover secrets before that image is sent. Keys, passwords, email addresses, phone numbers, card numbers, SSNs, IBANs, and the keywords you type are masked. Ordinary sentences stay visible. Everything in the shipped app runs on the machine. There is no cloud call.
QualcommExecuTorch
A
Lifeline LifeLine is an offline emergency-information assistant for situations where internet access is unavailable, such as remote hikes or network outages. Built for Snapdragon X Elite, it combines an ExecuTorch emergency-category classifier with a local Qwen conversational model through llama.cpp. Users can ask questions, browse stored medical references and access preparedness checklists. All AI processing stays on the device without cloud inference. LifeLine is a hackathon prototype for general information, not medical diagnosis or emergency dispatch.
QualcommQualcomm Snapdragon X EliteSamsung Galaxy S25 Ultra (PWA frontend/offline reference library)+14
JCZOffline Snapdragon Copilot+ PC A fully offline, explainable battery telemetry monitoring prototype for Windows on Snapdragon (ARM64). Uses only the Python standard library; no cloud, accounts, API keys, external JS, or model downloads. demo video
Qualcomm
R
Hakuna Matata!Creating a knowledge graph for project meetings using Snapdragon X elite with private on device speech transcription and summarizing.
Qualcomm
On-device AI that spots power-grid faults early and explains them in plain English. Runs offline on a Snapdragon Copilot+ PC.
QualcommPyTorchExecuTorch (XNNPACK)+5
On-device AI that spots power-grid faults early and explains them in plain English. Runs offline on a Snapdragon Copilot+ PC.
PyTorchExecuTorch (XNNPACK)ONNX Runtime+4
On-device AI that spots power-grid faults early and explains them in plain English. Runs offline on a Snapdragon Copilot+ PC.
PyTorchExecuTorch (XNNPACK)ONNX Runtime+4
