Overview
Sonder creates on-device vision models that learn how people work with their screens and act proactively. Its post-training, small vision-language models run directly on consumer laptops, observe user workflows, and convert them into shareable skills that teammates’ agents can reuse. This approach emphasizes local, privacy-conscious processing and enables enterprises to own their intelligence and data by keeping models and observations on their own devices. Compared with cloud-focused competitors, Sonder emphasizes on-device learning, workflow-to-skill translation, and collaboration through shared agent-enabled capabilities. The company’s goal is to empower teams to automate and accelerate work by turning observed screen workflows into reusable, enterprise-grade skills that live within the organization.
Work here?Claim your company
Simplify's take
What believers are saying
- Sonder claims 3× faster inference at half baseline size on laptops.
- Its October 26, 2026 website frames privacy and zero-friction individual adoption.
- The company is hiring a product engineer in New York, signaling active buildout.
What critics are saying
- Two founders and no public repositories leave Sonder exposed to execution failure.
- If consumer adoption stalls, enterprise learning-layer revenue never arrives.
- A data breach or workplace-monitoring scandal kills trust and ends the company.
What makes Sonder unique
- a16z speedrun SR007 backs Sonder’s consumer-laptop, on-device vision stack.
- Sonder’s models learn screen workflows, then package them as shareable teammate skills.
- It built a desktop RL environment in-house, resisting outside licensing interest.
Help us improve and share your feedback! Did you find this helpful?