unreallabsai/unreal-agent

What it solves

Unreal Agent is an async-first agent harness designed to provide a robust, durable, and composable framework for building AI agents. It focuses on solving the stability and reliability issues associated with agentic workflows, specifically handling input deduplication, session persistence, and the asynchronous execution of tools.

How it works

The system is built around a coordinator that manages the agent's lifecycle. It uses a Session inbox for input idempotency and a Session store to persist history and operation state, allowing for recovery and session forking. The process follows a specific flow: a Context builder assembles the model input, an LLM Adapter handles the provider communication, and a Tool registry manages the capabilities. When the LLM requests a tool, a Tool translator validates the call and converts it into serializable Operations that are executed by an Operation manager.

Who it’s for

Developers building complex AI agents that require high reliability, durable state management, and the process of separating tool definition from asynchronous execution.

Highlights

• Async-first architecture: Designed for non-blocking operations.
• Durable Sessions: Supports append-only persisted history that can be forked.
• Input Idempotency: Uses a session-scoped inbox to prevent duplicate processing of external inputs.
• Composable Components: Interfaces for the session store, LLM adapter, and operation manager are swappable to allow for custom implementations, such as remote sandboxing for tool execution.