prose

You can orchestrate parallel multi-agent AI workflows by loading a .prose program that uses a run-time VM-like environment to spawn sessions, route tasks, and merge results through context passing.

Yes, multi-agent workflows support context passing and discretionary evaluation, allowing you to coordinate sequential agent sessions and merge results dynamically within the orchestration program.

Multi-agent orchestration supports persistent state by leveraging pluggable state backends and memory persistence, which enables workflow resumption and auditing for long-running agent sessions.

You can run loops, conditionals, and reusable blocks for agent coordination by defining these patterns directly within a single .prose program executed by the orchestration VM.

The best way to manage state in parallel agent sessions is using pluggable state backends provided by the orchestration VM, ensuring reproducibility and reducing the need for ad-hoc scripting.

No external dependencies are required; the orchestration environment is self-contained, using the Task tool to spawn and coordinate agent sessions natively within the VM-like environment.