prose

Multi-agent workflow orchestration uses a VM-first model to coordinate sessions, manage state, and record evidence so long-running operations can resume and provide auditable proofs.

State management for persistent AI agents is handled through multiple backends including filesystem, in-context, SQLite, and PostgreSQL, allowing durable workflows to maintain memory and track execution state for cross-project collaboration.

Parallel execution for subagents is supported natively, enabling the VM to spawn sessions and coordinate concurrent workflows across multiple agents while maintaining memory and recording evidence for production-grade operations.

Running a minimal orchestration example involves executing a command like prose run examples/01-hello-world.prose, which demonstrates how the VM orchestrates a simple session and coordinates basic agent workflows.

Skill discovery in OpenProse workflows requires a SKILL.md file at the root directory and a defined frontmatter containing name and description, while optional resources in scripts/, references/, and assets/ load as directed by SKILL.md.

SQLite and PostgreSQL backends are fully supported for auditable AI workflow state tracking, providing durable storage that enables cross-project collaboration and verifiable evidence recording for long-running agent operations.