Google Open-Sources Harness with OpenCode & Codex Support
Google Research has open-sourced Harness², an agent framework designed to improve AI performance through recursive self-adjustment of components like prompts and memory. The project supports OpenCode and Codex and aims to enhance task execution without modifying model weights.
Why it matters
This framework offers a new approach to AI agent development, potentially allowing for more efficient and adaptable autonomous systems in research and practical applications.
The Google Research team has open-sourced an agent framework called Harness², which aims to enable AI Agents to achieve recursive capability improvement without modifying model weights, by continuously adjusting external components such as system prompts, memory, tool guidance, skills, and sub-agents.
At present, Harness² has been open-sourced on GitHub under the Apache 2.0 license, supports OpenCode and Codex, uses OpenCode by default, and can also be switched to Codex. Google Research also emphasized that Harness² is still a research project and is not an officially supported Google product.
GitHub address: https://github.com/google-research/harness2
The team splits this modifiable Harness into 8 types of components, including system prompts, guardrails, memory, tool guidance, sub-agents, Skills, scripts and plugins. These components jointly determine how the Agent constructs context, calls tools, delegates tasks, and completes work in multiple rounds of tool calls.
Get smarter about the news
Sign up free for a feed built around what you actually care about, Dive Deeper research on any story, and the full text of every article.
Create free accountAlready have an account? Sign in