Introduction

Everruns is a durable agentic harness engine built on Rust with a PostgreSQL-backed durable execution engine. It provides APIs for managing agents, sessions, and runs with streaming event output via SSE.

Overview

Everruns enables you to build reliable AI agents that can:

• Execute long-running tasks with durability guarantees
• Stream real-time events to clients
• Manage conversations through sessions
• Extend agent capabilities with modular tools

Key Concepts

Agents

Agents are AI assistants with configurable system prompts and capabilities. Each agent can be customized with:

• A system prompt that defines its behavior
• A set of capabilities that provide tools
• Model configuration for the underlying LLM

Sessions

Sessions represent conversations with an agent. Each session maintains:

• Conversation history
• Current execution state
• Configuration overrides

Capabilities

Capabilities are modular functionality units that extend agent behavior. They can:

• Add instructions to the system prompt
• Provide tools for the agent to use
• Modify execution behavior

See Capabilities for more details.

Getting Started

Quick Start

1. Deploy Everruns using the provided Docker images
2. Configure your LLM providers via the Settings UI
3. Create an agent with your desired configuration
4. Start sessions and interact through the API or UI

API Access

The API is available at your deployment URL:

• API Base: https://your-domain.com/api/v1/
• OpenAPI Spec: https://your-domain.com/api-doc/openapi.json

Architecture

Everruns uses a layered architecture:

• API Layer: HTTP endpoints (axum), SSE streaming
• Core Layer: Agent abstractions, capabilities, tools
• Worker Layer: Durable workflows for reliable execution
• Storage Layer: PostgreSQL with encrypted secrets and durable execution state

For detailed architecture information, see the GitHub repository.