Sim: Self-Hosted AI Agent Orchestration Platform
simstudioai/sim
Sim is a self-hostable workspace for building and running AI agents and workflows, built on Next.js, Bun, and PostgreSQL with a visual flow editor on ReactFlow. Reach for it if you want one place to wire agents into Slack, Notion, Salesforce, or a database instead of gluing together separate tools. Skip it if you just need one scripted agent — the setup wizard, Docker, and Kubernetes option are more infrastructure than that needs.
An Overview of Sim for AI Agent Management
Sim is an open-source AI workflow platform for building, deploying, and managing AI agents, acting as the coordination layer between a team and its agents. It ships as a hosted product at sim.ai or as a self-hosted app run with Bun and Docker. Agents, chat, workflows, tables, files, and a knowledge base share one workspace, so an agent can query structured data and pull from ingested documents without separate integrations.
Core Capabilities for AI Agent Orchestration
- ✓Broad LLM integration plus 1,000+ other integrations, per the README, with named connectors for Slack, Notion, HubSpot, Salesforce, and databases.
- ✓Three ways to build low-code AI agents: a visual flow editor built on ReactFlow, a conversational builder, or code directly.
- ✓A workspace-wide knowledge base plus file ingestion, so agents can read structured table data and documents without a separate vector store you wire up yourself.
- ✓A built-in table store ('Tables') works like a lightweight database inside the workspace, next to a shared file store the team and its agents both write to.
- ✓Workflow-activity monitoring — logs, schedules, and run history are tracked in the same workspace instead of a separate observability tool.
- ✓Local model support through Ollama and vLLM alongside hosted LLMs, so agents can run against a self-managed model instead of only calling a cloud API.
- ✓Chat runs as a Sim-managed service with its own API keys, created automatically during setup and manageable later from sim.ai/selfhost/settings/chat-keys.
Getting Started with Sim
Two paths. For the hosted product, sign up at sim.ai and skip local setup entirely. To self-host, clone the repo, then run `bun install` followed by `bun run setup`, which walks you through the whole install in one prompt: it sets up the database, creates secrets and your `.env` files, hooks up a Chat API key, then lets you choose whether Sim runs as Local dev, Docker Compose, or Kubernetes via Helm. Self-hosting requires Bun and Docker installed first. Once setup finishes, the app opens at http://localhost:3000. You can rerun `bun run setup <capability>` later — email, storage, sandbox, jobs, cache, knowledge, llm, or integration slack — to reconfigure one piece without repeating the whole wizard.
Strengths
- ✓One workspace covers agent building, a database-like table store, file storage, and a knowledge base, so you're not stitching together four separate tools for one workflow.
- ✓Three build modes — visual flow editor, conversational, and code — let you start fast and drop into code only where you actually need it.
- ✓Local model support via Ollama and vLLM means agents can run against models you host yourself, not just calls to a cloud LLM API.
- ✓The setup wizard handles the database, secrets, and .env generation in one step instead of leaving you to assemble the config by hand.
- ✓Apache-2.0 licensing keeps self-hosted use unrestricted.
Considerations Before Adopting Sim
- △Self-hosting requires both Bun and Docker installed before you start — there's no lighter single-binary option described in the README.
- △Running a bare `sim` command needs `bun link`, and on a Homebrew-installed Bun that binary lands in `~/.bun/bin`, which isn't on PATH by default, so you have to export it yourself.
- △Chat is a Sim-managed service tied to an API key generated through sim.ai during setup, so part of the self-hosted stack still depends on Sim's own hosted infrastructure.
- △The README doesn't document a plugin count, benchmark numbers, or a version history, so you can't gauge release cadence or maturity from the repo page alone.
Alternatives for AI Agent and Workflow Management
Frequently Asked Questions about Sim
Sim is licensed under Apache License 2.0, so self-hosting, modifying, and redistributing it stay within Apache-2.0 terms.
Self-hosting Sim requires Bun and Docker installed locally; the setup wizard then sets up the database for you and lets you pick Local dev, Docker Compose, or Kubernetes via Helm as the run mode.
Sim's README states it connects to 1,000+ integrations and every major LLM, with named connectors for Slack, Notion, HubSpot, Salesforce, and databases.
Sim supports local models through Ollama and vLLM, so agents can run against a self-hosted model instead of a cloud LLM API.
Sim runs on Next.js (App Router) and Bun, stores data in PostgreSQL through Drizzle ORM, handles auth with Better Auth, and validates schemas with Zod; the UI uses Shadcn and Tailwind CSS, with ReactFlow powering the visual flow editor.
The problem it solves
Teams that want an AI agent to actually do something — read a Notion page, write to Salesforce, query a database — usually end up wiring together a workflow tool, a vector store, a file store, and a chat UI as separate pieces. Sim's approach is to put agents, a table store, file storage, and a knowledge base in one workspace, so a single workflow can touch all of them without custom glue code between services.
Best use cases
- •Wiring an internal support agent for AI workforce automation that reads a Notion knowledge base and writes tickets into Salesforce or HubSpot, skipping custom API glue code.
- •Running a self-hosted agent workflow on data that can't leave your infrastructure, using local models through Ollama or vLLM instead of an external LLM API.
- •Standing up a team chat surface backed by workflow automation that queries a Postgres-backed table store and files in the same workspace.
- •Prototyping a no-code AI agent conversationally or visually first, then dropping into code for the parts that need custom logic.
Who should try it — and who should skip
Try Sim if you're a small team that wants one self-hosted workspace to build agents, store the data they touch, and hook them into tools like Slack, Notion, or Salesforce without running four separate services. Skip it if you need a single lightweight agent script — the Bun-plus-Docker setup, database provisioning, and Kubernetes option are more infrastructure than a one-off automation needs. Also skip it if you want to stay fully independent of Sim's own hosted service, since Chat features route through a Sim-managed API key even in a self-hosted deployment.