Designing an event-driven architecture for an autonomous companion robot
View Anya on GitHub↗Executive Summary
A robot is fundamentally different from a conventional web application. It does not simply wait for a request and produce a response. It is continuously receiving information from its environment while simultaneously processing perception, memory, cognition, speech, behavior, and hardware events. Anya was designed as an experiment in embodied intelligence around that constraint. Instead of creating one monolithic application, we designed the system as a collection of independent but coordinated subsystems that communicate through an event-driven architecture.
The Challenge
The initial problem appears simple: sensors provide information, an AI system decides what to do, and the robot performs an action. The architecture becomes much harder once multiple capabilities operate concurrently. Vision should not stop because the robot is speaking. Memory should persist independently of a single interaction. A cognitive process may take significantly longer than an immediate physical reaction. Hardware can introduce latency and unreliable communication. Tightly coupling all of these workloads into one process would make the system difficult to extend and difficult to reason about. Anya therefore needed clear boundaries between perception, cognition, memory, reflexes, habits, expression, and hardware interfaces while still allowing those components to respond to the same stream of events.
The Approach
We designed Anya around an event-driven architecture using NATS as the communication layer between independent subsystems. Perception can publish observations without directly depending on the cognition system. Memory can independently determine what information should persist. Cognition can consume relevant events and produce decisions. Reflexes can respond to immediate conditions without unnecessarily invoking a large reasoning model, while habits can handle recurring behavior independently of deliberate cognition. Expression systems then translate resulting decisions into speech, behavior, and interaction. This architecture also creates a boundary between the cognitive software and physical hardware, making it possible to evolve individual components without rewriting the entire robot. The design was developed with real-world constraints in mind, including limited compute, wireless communication, sensor latency, external devices, and the need for different workloads to continue operating concurrently.
The Impact
The result is a modular foundation for experimenting with embodied intelligence without tying the entire robot to one process, model, or hardware configuration. The event-driven design makes the system easier to extend because new capabilities can participate in the same communication architecture rather than creating another tightly coupled dependency. More importantly, Anya demonstrates a systems-engineering approach to AI: intelligence is not treated as one model invocation, but as a collection of perception, memory, reasoning, communication, and action systems that have to operate together in the physical world.
What’s sitting on your engineering backlog?
Give us one task. We’ll take it from scope to production while your team keeps building the core product.