Versions:
CrewHaus is a modular meta-harness developed and published by CrewHaus, designed to simplify the way developers define and deploy agent-based systems. Its core purpose is to compile a single high-level harness specification, expressed in a configuration file called crewhaus.yaml, into multiple distinct agent runtimes, eliminating the need to author and maintain separate configurations for each deployment target. This approach positions CrewHaus within the developer tooling and agent framework category, where consistency, reuse, and portability of agent definitions are ongoing challenges. The software is particularly relevant for teams and individuals building with large language model–driven agents who want one authoritative source of truth for their agent behavior. From one spec, CrewHaus can generate a wide range of runtime targets, including a CLI agent for terminal-based interaction, a channel bot for chat or messaging integrations, a RAG pipeline for retrieval-augmented generation workflows, a multi-agent crew for coordinated collaboration between multiple agents, and an eval harness for testing and evaluating agent performance. It also supports more specialized targets such as a voice or realtime agent for spoken, low-latency interactions and a browser or computer-use agent capable of operating graphical or web-based environments, with additional targets available beyond these examples. Typical use cases include prototyping an agent once and then shipping it across several surfaces, maintaining parity between an agent's evaluation setup and its production behavior, and iterating quickly on a specification without rewriting runtime-specific glue code. By treating the harness spec as a compilation artifact rather than a fixed integration, CrewHaus enables experimentation across runtime types while keeping the underlying definition stable. The current release is version 0.6.0, and the software has seen five published versions to date, reflecting an active development history. As a pre-1.0 tool, it remains in an early stage of its lifecycle while already covering a broad spread of agent runtime categories.
Tags: