// the find
simstudioai/sim
Build, deploy, and orchestrate AI agents. Sim is the central intelligence layer for your AI workforce.
Sim is a visual AI agent builder — think n8n but LLM-first, with a canvas workflow editor, a chat interface that can build workflows from natural language ("Mothership"), built-in knowledge bases, and a structured database layer. It targets teams who want to wire up AI automations across 1000+ integrations without writing most of the glue code. Self-hostable via Docker or a one-liner npx command.
The technical stack is well-chosen for what this does: Drizzle ORM gives type-safe queries, pgvector handles semantic search for the knowledge base without a separate vector store, isolated-vm gives sandboxed JS execution for user-defined functions without spinning up containers per call, and E2B handles heavier remote code execution. The dogfooding is real — they maintain `.agents/skills/` rules and `.claude/rules/` that govern how AI assists in their own development, which means the patterns they expose to users are tested on their own codebase. The deployment story is actually good: `npx simstudio` wraps Docker Compose and gets you running in one command, and the manual setup docs are thorough enough to follow without guessing. The Mothership interface building and wiring workflows from plain language is the genuinely interesting bet here — most visual workflow tools stop at "drag blocks," this one tries to let you describe intent and have the canvas update.
Copilot — the natural-language workflow builder, which is the main differentiator — is a Sim-managed cloud service even on self-hosted instances. You still need a sim.ai API key. If that service goes away or changes pricing, self-hosters lose the headline feature. The Docker setup requires 12GB RAM, which rules out cheap VPS hosts for teams that want to self-host cheaply and casually. The integration surface is enormous (100+ documented tools in the repo) and the maintenance burden scales with it — every third-party API that changes its auth scheme or response shape is a bug waiting to land. There's no visible story in the README or docs around workflow versioning, rollback, or audit logging for teams that care about reproducibility; at this scale of integrations that's a real gap for anything beyond internal tooling.