prose

You orchestrate multi-agent AI workflows by using a unified VM to spawn subagents via the Task tool, enabling parallel or sequential execution. The runtime coordinates session composition, context passing, and program imports to execute complex pipelines end-to-end.

Yes, you can use SQLite, PostgreSQL, or filesystem backends for persistent agent memory in orchestration. These state-management primitives support long-running runs by maintaining reliable state and traceability across multiple agent sessions.

You build modular AI workflows by defining agent templates and sessions in a .prose file, then running it with the VM. This approach supports context passing and nested imports, allowing you to compose modular pipelines across single-run or cross-project contexts.

Yes, the orchestration runtime supports cross-project context passing through nested program imports and session composition. It applies to building complex pipelines and captain's-chair patterns that span multiple project contexts reliably.

You need a unified VM for orchestrating AI agents when building production-grade workflows that require reliability and traceability. It centralizes the runtime and state-management primitives needed for complex patterns spanning multiple agents and sessions.