Sim Wants to Be the Figma for Your AI Agent Team

What it does

Sim is an open-source platform for building and running AI agent workflows through a visual node editor. You arrange agents, tools, and logic blocks on a canvas—built with ReactFlow—and execute them against various LLMs and vector databases. It packages the whole thing as a Next.js application that can be cloud-hosted or run on your own hardware, with support for local models via Ollama and vLLM.

The interesting bit

The project is refreshingly upfront about being a well-architected integration: ReactFlow, Drizzle ORM, Better Auth, and Trigger.dev stitched together inside a Turborepo monorepo. That makes it competent glue rather than reinvention. The standout detail is sandboxed execution—untrusted code runs inside isolated-vm and E2B—so the visual editor does not become a remote-code-execution footgun.

Key highlights

• Visual workflow builder using ReactFlow; demos show instant execution on the canvas.
• Claims to connect 1,000+ integrations and LLMs, though the README never names them.
• Built-in RAG support through PostgreSQL with pgvector for document grounding.
• Self-hostable via Docker or manual setup using Bun, Node 20+, and PostgreSQL.
• Sandboxed code execution using isolated-vm and E2B.

Caveats

• The Copilot assistant is a Sim-managed service; self-hosted instances must still fetch an API key from the cloud-hosted sim.ai to use it.
• The README advertises “1,000+ integrations” without listing providers or detailing how they are implemented.

Verdict

Worth a look if you want a self-hostable, visual node editor for agent workflows and do not mind a standard TypeScript stack. Give it a pass if you need a fully offline solution with no cloud dependencies, or if you prefer code-first orchestration over clicking nodes.