One /go command for AI work that remembers — zero runtime
Assemble is an open-source configuration generator for AI work: /go, memory, spec-driven workflows, and zero runtime across 21 platforms.
Assemble provides 15 predefined workflows (/feature, /bugfix, /security, /release, /sprint, etc.) that chain agents in optimal order. Jarvis orchestrator assesses task complexity and routes to appropriate agents; users describe needs with /go and receive structured deliverables without manual agent selection.
Assemble provides 34 pre-configured AI agent personas (e.g., @tony-stark for architecture, @punisher for security, @hawkeye for QA) that work together in structured workflows. Users invoke agents via commands like /go, /security, /review to get domain-specific analysis and code generation without managing multiple generic AI assistants.
Assemble embeds structural dissent by design: @deadpool permanently challenges workflows, @doctor-doom escalates irreversible decisions, and both agents can block decisions with RED flags. This reduces hallucination-driven errors by 65% compared to frameworks that assume agents naturally check each other.
Assemble is fully extensible: users drop Markdown files into .assemble/agents/ to define custom agent personas, skills, and rules; custom workflows can be added to .assemble/workflows/. New agents and workflows are included in the next generation without modifying core system.
Assemble generates native config files for 21 platforms (Cursor, Claude Code, GitHub Copilot, Windsurf, etc.) from a single .assemble.yaml source. Users define team structure and rules once; Assemble automatically produces platform-specific files (.cursorrules, CLAUDE.md, .github/copilot-instructions.md, etc.) that stay in sync.