Vega provides the runtime layer between high-level intent and robot-specific functionality. Instead of requiring users or applications to operate individual robot subsystems directly, Vega coordinates available Capabilities around a clearly defined outcome.
The problem Vega addresses
Robots may have powerful individual functions, but those functions are often fragmented across vendor tools, hardware interfaces, scripts, and operating procedures. Building a useful robot experience therefore requires more than connecting an AI model to a collection of commands.
Real-world robot tasks are stateful and long-running. They may be interrupted by process failures, network loss, robot restarts, unavailable services, or changing environmental conditions. Without a persistent runtime, the system may lose task context, repeat completed work, or fail to continue after an interruption.
Vega provides a common runtime for managing this execution lifecycle.
How Vega works
Vega sits between people or applications and the Capabilities available in a robot environment.
- A person or application describes an intended outcome.
- Vega interprets the request in the context of the current robot, environment, and available Capabilities.
- Vega creates and manages an executable task.
- Capability Providers perform bounded robot- or service-specific operations.
- Vega tracks execution state, results, failures, and recovery throughout the task lifecycle.
This separation allows Capability developers to focus on specific functions while Vega manages the broader embodied-agent experience.
Core features
Vega preserves the context and state required to support ongoing robot work across multiple interactions and execution stages.
Vega tracks task progress and execution state so multi-step work can continue coherently rather than as disconnected commands.
When a process, Capability, connection, or robot is interrupted, Vega uses persisted state and results to determine how to resume, retry, or fail safely.
Vega discovers, invokes, and coordinates modular Capabilities according to the task, environment, and declared Capability contracts.
Capabilities
A Capability is a bounded, machine-readable unit of functionality that Vega can discover and invoke.
A Capability declares:
- What it does
- Its required inputs
- Its expected outputs
- Its runtime and hardware requirements
- Its supported execution behavior
- Its failure conditions and limitations
Capabilities may interact with robot hardware, ROS 2 systems, sensors, external services, models, or other software components. Capability contracts allow different robots and deployments to expose different functionality while maintaining a consistent integration model.
The Vega ecosystem
| Component | Role |
|---|---|
| Vega Runtime | The proprietary runtime responsible for task execution, state, memory, orchestration, and recovery. |
| Vega SDK | The open-source development toolkit for creating and integrating Vega-compatible components. |
| VCP | The open communication protocol used between Vega, Capability Providers, robots, and external systems. |
| Vega Capabilities | Open-source or privately distributed packages that provide specific robot and service functionality. |
Vega Runtime is distributed as a free installable package. Developers do not need access to Vega's proprietary implementation to build compatible Capabilities or VCP integrations. The SDK, VCP specification, and reference Capabilities provide the public collaboration surface for the ecosystem.