OpenAI fully open sourced Codex Harness, and further opened up the underlying capabilities of AI programming agents. OpenAI recently further opened up the core capabilities related to Codex Harness, and opened the intelligent agent operating framework that drives different Codex product forms to developers.

OpenAI fully open-sources Codex Harness, further opening up the underlying capabilities of AI programming agents. OpenAI has recently further opened up the core capabilities related to Codex Harness, opening the agent operating framework that drives different Codex product forms to developers. Codex Harness is the core Agent operating environment behind Codex. It is responsible for connecting models, users and tools, and managing processes such as task execution, context, session and code operations. OpenAI has previously open sourced the Codex CLI and related core code, and continues to improve Harness and App Server capabilities. From an architectural point of view, Codex Harness is not a simple code generation tool, but a complete Agent execution framework. Its core includes Agent Loop, thread life cycle and persistence, configuration and identity authentication, tool execution and expansion mechanism, etc. Through the Codex App Server, developers can use the bidirectional JSON-RPC protocol to connect Harness to different clients and obtain real-time event streams to build IDEs, desktop applications and other customized agent products. Currently, Codex Harness supports multiple usage scenarios such as Codex Web, CLI, IDE extensions and macOS applications. OpenAI also previously publicly shared the Harness engineering practice of "all code written by Codex": an internal product formed about 1 million lines of code in about 5 months, and completed about 1,500 Pull Requests. The core concept of the team shifted from "manual writing of code" to "human beings set goals and agents execute". The further opening of Harness this time means that AI programming competition is extending from model capabilities to Agent runtime, tool invocation, context management and engineering feedback closed loop. As the underlying Harness gradually becomes standardized, developers can build richer professional AI programming and software engineering workflows based on Codex's core Agent capabilities. via AI News (author: AI Base)