OpenRath

OpenRath is an open source, session first runtime for building multi agent workflows. It is developed by Rath Team, which builds infrastructure for agent systems in which tool calls, sandbox execution, branching, memory and validation all remain inspectable. The starting point is a criticism of systems that reduce agent work to a chat transcript, which makes long running tasks hard to resume, audit, branch or compare. OpenRath makes the session the unit of work. A session carries messages, tool evidence, runtime placement, validation results, memory reads and workflow lineage throughout the process. The project explores session graphs, workflow composition with Python first modules and explicit handoffs, and sandbox backed execution. The stated goal is to preserve the path of work rather than just the final answer, so that builders can inspect, replay, branch and compare agent runs. Reference examples cover trading, engineering and research agents. The team describes OpenRath as a runtime direction rather than a finished platform, so it is best suited to engineers and researchers building their own agent systems who are comfortable reading code. The site links to documentation with quickstarts and an API reference, a GitHub repository, a PyPI package, example workflows and a blog of research notes and engineering updates. The site does not publish pricing. The package is distributed through PyPI and the source is public on GitHub.