vvm

VVM (Vibe Virtual Machine) is a domain-specific language (DSL) designed for building, orchestrating, and executing agentic workflows where the LLM acts as the primary runtime. It provides developers with explicit control over agent boundaries, semantic flow, and concurrency, making it ideal for creating complex, reliable AI systems that require multi-step reasoning, state persistence, and error handling. By using VVM, users can replace monolithic agent scripts with modular, reusable components that define clear contracts for input, processing, and output.

• Define specialized agents with custom prompts, model configurations, and permissions to ensure focus and cost efficiency.

• Implement complex control flows using semantic predicates, pattern matching with match statements, and AI-driven branching via choose instructions.

• Leverage explicit concurrency with parallel map operations (pmap) and support for scalable state backends including in-context, filesystem, SQLite, and PostgreSQL.

• Create sophisticated agent memory systems using digest and ledger mechanics to maintain persistent, inspectable, and multi-tenant state across workflow executions.

• Utilize a comprehensive command-line interface for development, including /vvm-boot to initialize projects, /vvm-compile for validation, and /vvm-run to execute workflows.

• Maintain high-quality automations through constraint-based requirements and iterative refinement loops that verify output quality.

• The skill is intended for AI engineers and developers building autonomous agents, research pipelines, or iterative coding workflows. It works best when tasks require distinct logical stages or collaborative agent behavior.

• Expected inputs include .vvm files containing agent logic, while outputs consist of structured execution results, data artifacts, or side-effect operations in the host system.

• Practical constraints: VVM is currently optimized for agent orchestration; users should manage permissions and review gates for high-impact or sensitive operations.

• When building, consult the provided spec.md for language syntax, patterns.md for design best practices, and antipatterns.md to avoid common pitfalls like 'God Agents' or context explosion.