# A better way to build a robotics stack

What custom robotics tooling costs at scale

## 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.