A better way to build a robotics stack

A custom stack doesn't include the hard parts

For most robotics startups, building on a custom stack is the default path. The tooling is often free, flexible, and battle-tested enough to get to a prototype. So teams build their own OTA pipelines, their own data infrastructure, their own deployment tooling.

The problem shows up later. OTA updates, data pipelines, fleet management, security — none of it comes with the stack. Each one has to be built separately — its own engineering project, running in parallel with the actual product you're trying to ship. By the time you're ready to scale, a significant chunk of your engineering bandwidth is maintaining infrastructure that doesn't differentiate you at all. That's the tradeoff Viam is built to eliminate.

Building a better solution

The table below maps the most common custom stack challenges to how Viam handles them. Early-stage founders will recognize the Development Velocity challenges: hardware, vision, dev environment. Scaling founders will know the Production Operations pain points, like OTA, fleet management, security.

‍

Feature	Custom stack	Viam
Development velocity
Hardware integration	Swapping a component means a new driver, config changes, and retesting the whole integration.	Change a line in your config to change hardware. Viam pulls the driver, your code stays the same.
Vision & perception	Setting up a CV pipeline means writing boilerplate for every detection task — object recognition, color detection, and point clouds each require separate integration work.	Vision service built in. Object detection, color detection, and point cloud support available out of the box via a clean API.
Developer environment	Custom stacks carry their own environment requirements — distro pinning, OS versions, and hardware compatibility you have to sort out before writing any code.	One install command on Linux, macOS, or Windows. Runs on the ARM hardware your robots ship on.
SDKs & language support	Custom stacks come with their own abstractions you have to learn before the robot does anything.	Python, Go, TypeScript, C++, Flutter, and more. If you can call a function, you can control a robot.
Intelligent autonomy
AI/ML deployment	Your model is trained — but getting it on the robot means resolving CUDA conflicts, managing dependencies, and wiring an inference pipeline from scratch.	Grab a model from the Registry or upload your own. One API call, running on-device.
Motion & pathing	Without a motion planning layer, obstacle avoidance and path planning mean tuning your own controllers and handling edge cases yourself.	Built-in motion service handles path planning and obstacle avoidance. Define a destination, Viam figures out how to get there.
Data pipeline	Capturing data works, getting it into a training pipeline or querying it across machines without building the plumbing yourself doesn't.	Configure capture once. Viam syncs, stores, and makes your data queryable automatically.
Production operations
OTA updates	OTA updates aren't included — every team builds their own, which means owning that infrastructure on top of everything else.	Push from the CLI, robots pull on schedule. Roll back from the dashboard if something breaks.
Observability & debugging	Production observability — aggregated logs, metrics, fleet alerts — requires integrating third-party tooling yourself.	Logging, live log streaming, and machine alerts built-in. No separate tools to set up.
Fleet scaling	The scripts and manual configs that got your prototype working don't scale to a fleet. Most teams rebuild from scratch.	Prototype config is your production config. Apply it to hundreds of machines. No rewrites required.
Security & compliance	No built-in auth, encrypted communication, or compliance framework — each one is a separate integration project before your first enterprise conversation.	Encrypted connections, SOC 2 Type II, role-based access control, and API keys built in from day one.

‍

Viam doesn't make the hard parts of robotics disappear — the hardware is still complex, the edge cases are still real, and building a great product still takes great engineers. What it does is make sure your team is spending their time on the problems that are actually yours to solve.

A better way to build a robotics stack

A custom stack doesn't include the hard parts

Building a better solution

Find us at our next event