Skip to content

Architecture Overview

RTC Agent’s architecture consists of three parts: Browser / Frontend (virtual file system + Web Component), RTC Agent Server (Agent engine + context management + real-time communication), and LLM Provider (AI reasoning).

Function Registry: Function metadata (name, parameter Schema, description) registered by developers via agentConfig or defineRegistry, stored as Markdown files in the /functions/ directory of the virtual file system. The Agent engine discovers available business capabilities by reading these files. See Skill System.

PrincipleImplementationUser Value
🔐 Privacy FirstFile operations execute on the frontend; data stored in IndexedDBSensitive data is never uploaded to the server
🧩 File System as InterfaceAI operates business logic via ls/read/write/grepDevelopers only need to maintain Function documentation
Real-Time SyncCentrifuge dual-channel pushUsers can see every step the AI takes
♻️ Resilient RecoveryRedis Checkpoint + 100% delivery guaranteeNetwork disconnections and restarts are seamless

Why “File System as Interface”? LLMs naturally understand file operations (ls, cat, grep) — no need to learn custom APIs. Developers just write a Markdown document for each Function, and the Agent can automatically discover and compose calls — lower integration cost and higher observability compared to traditional RPC registration.

LayerTechnologyPurpose
BackendGo 1.27Main service language
DatabasePostgreSQL + GORMPersistent storage
CacheRedisCheckpoint, streaming buffer, Pub/Sub
Real-Time CommunicationCentrifuge WebSocketBidirectional push
AI FrameworkAnthropic SDK + EinoAgent engine
FrontendLit Web ComponentsComponentized UI
ObservabilityOpenTelemetry + Jaeger + PrometheusTracing + Metrics