Architecture
Everruns is a headless durable agentic harness engine built for reliability and scale. It provides a REST API for managing agents, sessions, and runs with real-time event streaming via SSE.
Platform Overview
Section titled “Platform Overview”
Key Design Principles
Section titled “Key Design Principles”| Principle | Description |
|---|---|
| Headless / API-First | Integrate via REST API. No UI required for production. |
| Agentic Loop | Core pattern: Reason → Act, repeated until task complete. |
| Durable Execution | Agent state survives restarts. Never lose progress. |
| Horizontal Scaling | Add workers to increase throughput. |
| Provider Agnostic | OpenAI, Anthropic, Gemini, or custom LLM providers. |
Components
Section titled “Components”For the deployable shape of these components, which run as separate processes, which are required, and which are optional infrastructure like NATS and Valkey, see Physical Architecture.
Control Plane
Section titled “Control Plane”Central coordinator that exposes the REST API and manages all state in PostgreSQL:
- Agents - AI assistant configurations
- Sessions - Conversation state and history
- Events - Real-time event streaming via SSE
Workers
Section titled “Workers”Stateless executors that run agentic loops composed from atoms (Input → Reason → Act). Workers are:
- Scalable - Add more to handle concurrent sessions
- Fault-tolerant - Failed tasks automatically recovered
- Stateless - All state lives in PostgreSQL
Management UI (Optional)
Section titled “Management UI (Optional)”Administrative interface for platform operators. Not required for production use.
- Agent configuration
- Session monitoring
- LLM provider settings
See Management UI for details.
Further Reading
Section titled “Further Reading”- Physical Architecture - The deployable components: PostgreSQL, NATS, Valkey, workers, and how they connect
- Introduction - Getting started
- API Reference - Full API documentation
- Capabilities - Extend agent functionality