What is Sim?
Sim is a workspace for building AI agent workflows. You draw them on a canvas of blocks (agents, routers, conditions, loops, human approval, API calls), describe them in a chat that builds the workflow for you, or write code. Workflows deploy as APIs, chat pages or MCP servers, and sit next to built-in tables, file storage and knowledge bases.
The README counts 280+ integrations with 5,800+ actions and 530+ triggers. Agent blocks call models from OpenAI, Anthropic, Google, xAI, Groq, DeepSeek, Bedrock, Azure, Vertex and OpenRouter, or local models through Ollama and vLLM.
The whole repository is Apache 2.0. Sim sells a hosted cloud and an Enterprise plan, and ships a macOS desktop app. The repository dates from January 2025, is still pre-1.0, and v0.9.62 shipped on 17 November 2026.
Is Sim worth it?
Sim is the agent builder to look at if you want a visual canvas like n8n's but a real Apache 2.0 licence with no commercial carve-outs. The repository has about 30,200 GitHub stars and 45 releases in the month to 17 November 2026, when v0.9.62 shipped. That pace brings features quickly and leaves a self-hoster with a moving target: version numbers are still below 1.0, and the compose file moves the app, realtime and migration images in lockstep because they share a database schema.
Self-hosting is less self-contained than the licence implies. The Chat assistant that builds workflows from a description is a Sim-managed service: a self-hosted install needs a COPILOT_API_KEY from sim.ai, or you hide Chat. Plain JavaScript runs in isolated-vm inside the app container, but Python, shell and JavaScript with imports need a remote sandbox from E2B or Daytona. Knowledge bases need an embedding key from OpenAI, Azure, OpenRouter or Gemini, or your own Ollama.
On Sim Cloud every run costs at least one credit, worth $0.005, and model calls through Sim's keys are charged on top at Sim's rates; bringing your own keys removes the model charge but not the base fee. Free gives 1,000 credits once, 2 concurrent runs and a 5-minute sync timeout. Pro at $25 per user a month includes 6,000 credits and 50 concurrent runs; Max at $100 includes 25,000. SSO, access control and SOC 2 paperwork are Enterprise only on the cloud, while the self-hosting docs switch the same features on with environment variables.
The catch: the Docker defaults trust whoever finds the server. Email sign-up is open, verification is not enforced without a mail provider, and the compose file publishes the app on port 3000 and the realtime service on port 3002 on every interface, with no TLS. The docs' own checklist asks you to restrict sign-up, add TLS and back up ENCRYPTION_KEY separately, since losing it makes stored credentials unreadable.
Our take: Sim suits a team that wants an open licence and a polished canvas and is willing to pin a version and patch often. For agent-heavy RAG apps Dify is more settled, and n8n still has the deeper bench of business integrations.
Where it earns its place
A three-person customer-support team at a SaaS start-up
Triage inbound tickets and draft replies from the help centre
A webhook trigger feeds an Agent block that reads a knowledge base built from the help docs, classifies urgency and drafts an answer for a person to approve. On Pro at $25 per user a month, each run costs a base credit of $0.005 plus model usage unless you bring your own key.
An IT team at a European insurer that cannot send customer data to a US SaaS
Run agent workflows against internal systems on its own servers
Self-host with npx sim-setup on a machine with 4 cores and 16 GiB of RAM, the docs' team-sized example. Point Agent blocks at Ollama or vLLM, hide Chat to avoid Sim's service, and enable SSO and audit logs with the enterprise environment flags.
A developer using Claude Code
Run and deploy workflows from the coding agent instead of the browser
Add Sim's hosted MCP server at https://mcp.sim.ai/mcp and sign in with OAuth; the agent can manage workspaces, run and deploy workflows and read logs. Self-hosted instances expose their own MCP URL.
Before you commit
- Setup
- Needs Node.js 20+, Bun and Docker Compose 2.24+. npx sim-setup generates the secrets, writes .env, connects a Chat key and starts the published images; open http://localhost:3000. Sizing examples run from 2 cores and 12 GiB of RAM for evaluation to 8+ cores and 32+ GiB for larger workloads; Helm charts cover Kubernetes.
- Integrations
- 280+ integrations with 5,800+ actions and 530+ triggers per the README; workflows deploy as REST APIs, chat pages or MCP servers; a hosted MCP server, CLI and API for driving Sim from Claude, Claude Code or Codex.
- Data and privacy
- Self-hosted data stays in your Postgres (pgvector) and Redis; OAuth tokens and stored keys are encrypted with ENCRYPTION_KEY. Chat sends requests to Sim's service when enabled. Docker and Helm set NEXT_TELEMETRY_DISABLED=1 by default. Sim Cloud cites SOC 2 Type II, ISO 27001 and GDPR in its Trust Center.
- Cost at scale
- Self-hosted with billing disabled has no Sim charges, only servers, model APIs and any E2B or Daytona sandbox. On Sim Cloud each run costs at least $0.005 plus model and hosted-tool charges; Max is $100 per user a month for 25,000 credits and 200 concurrent runs.
- Limits
- 329 open issues and pull requests. Cloud Free allows 2 concurrent runs, 5 tables and 30 days of logs; Pro and Max cap sync runs at 50 minutes and async runs at 90 minutes. The production compose file caps the app container at 8 GB of memory and does not mount the upload directory.
What’s good
- Whole codebase under Apache 2.0, including enterprise features
- Build by canvas, by chat or in code; deploy as API, chat page or MCP server
- Broad model support, including Bedrock, Vertex, Ollama and vLLM
- Built-in tables, files and knowledge bases next to workflows
What’s not
- Pre-1.0 with dozens of releases a month
- Chat builder on self-hosted installs needs a key from Sim's service
- Python and shell code steps need an E2B or Daytona account
- Open sign-up and ports 3000 and 3002 exposed by default
Best for
Product and operations teams that want to build AI agent workflows visually under a permissive licence, on Sim Cloud or their own servers.
Sim alternatives
Sim: common questions
- Is Sim worth it?
- Sim makes agent workflows quick to draw and its licence is genuinely open, but the project moves too fast and leans on its own cloud in too many places for a calm production install. We rate it 3.9 out of 5.
- How much does Sim cost?
- Apache 2.0, free to self-host. Sim Cloud: Free with 1,000 one-time credits; Pro $25/user/month (6,000 credits); Max $100/user/month (25,000 credits); Enterprise custom. 1 credit = $0.005; annual billing 15% off.
- What are the downsides of Sim?
- Pre-1.0 with dozens of releases a month; chat builder on self-hosted installs needs a key from Sim's service; python and shell code steps need an E2B or Daytona account; open sign-up and ports 3000 and 3002 exposed by default.
- Who is Sim best for?
- Product and operations teams that want to build AI agent workflows visually under a permissive licence, on Sim Cloud or their own servers.
- How much does Sim cost for a team or in production?
- Self-hosted with billing disabled has no Sim charges, only servers, model APIs and any E2B or Daytona sandbox. On Sim Cloud each run costs at least $0.005 plus model and hosted-tool charges; Max is $100 per user a month for 25,000 credits and 200 concurrent runs.
- Is Sim safe for company data?
- Self-hosted data stays in your Postgres (pgvector) and Redis; OAuth tokens and stored keys are encrypted with ENCRYPTION_KEY. Chat sends requests to Sim's service when enabled. Docker and Helm set NEXT_TELEMETRY_DISABLED=1 by default. Sim Cloud cites SOC 2 Type II, ISO 27001 and GDPR in its Trust Center.
- What are the best Sim alternatives?
- The closest alternatives we have reviewed are n8n, Dify, Langflow and Activepieces.
Answers checked Nov 18, 2026.