The source code for Claude Code, a project developed by Anthropic, has reportedly been leaked online. Instead of reacting negatively, a former Anthropic product manager has leveraged this incident to create a new open-source framework. The former manager analyzed the architecture of Anthropic's multi-agent orchestration system, which includes a coordinator for decomposing goals into tasks, a command system, a message bus, and a task scheduler considering dependencies. He then rebuilt these patterns from scratch as a separate open-source framework, without directly using Anthropic's leaked code. The result is the independent open-source framework named 'open-multi-agent'.
A significant distinction between 'open-multi-agent' and the original claude-agent-sdk lies in their execution approaches. The original SDK launched a separate command-line interface (CLI) process for each agent, which was resource-intensive. In contrast, 'open-multi-agent' operates in-process, meaning all agents run within the same process. This streamlined approach makes the framework remarkably versatile. It can be deployed across various environments, including serverless functions, Docker containers, and CI/CD pipelines.
Curiously, a data leak, typically viewed as a major setback, has acted as a catalyst in this instance. Anthropic's architectural innovations, even before their official release, proved valuable enough for someone to replicate and share them with the open-source community. This situation underscores how significant ideas can emerge outside formal channels. Businesses now have an additional, more accessible tool for experimenting with AI agents.