Sim

What it is

Sim is a visual builder for AI agents and the workflows around them, and the core is licensed Apache-2.0 at github.com/simstudioai/sim. You draw agent blocks, tool calls, conditions and loops on a canvas, deploy the result as an API, a chat, a schedule or a webhook, and read every run’s trace in the logs. Since v0.6 on 17 March 2026 the product also carries Mothership, a natural-language control plane that builds and edits workflows, tables, files and knowledge bases from one conversation. That launch is when Sim stopped calling itself a workflow builder and started calling itself an AI workspace.

MCP runs in both directions. Agents call any external MCP server you register in the workspace, and a deployed workflow can be published as an MCP tool that Claude Code, Claude Desktop, Cursor or Codex call directly. The repository README claims 1,000+ integrations, including Slack, HubSpot, Salesforce and Notion. Local models run through Ollama or vLLM.

The company was founded in 2025 by two UC Berkeley alumni, went through Y Combinator, and raised a $7M Series A led by Standard Capital with Perplexity Fund and SV Angel participating. The repo sits near 30,000 GitHub stars and ships several releases a week — v0.9.1 to v0.9.3 landed within two days in late September 2026.

The fork it resolves

The question this page answers: n8n’s licence is a problem for us — what is a genuinely open-source automation engine we can self-host?

n8n ships under the Sustainable Use License, which limits use to internal business or non-commercial purposes. That is fine for most ops teams and a wall for anyone who runs workflows for paying customers. “Show HN: Sim – Apache-2.0 n8n alternative” ran on 11 December 2025 at 240 points and 61 comments, and the licence was the thread’s main argument. Apache-2.0 lets you modify, embed and resell the core.

The second half of the fork is what you are automating. n8n and Zapier started as integration plumbing and added agents later. Sim started from the agent block. If most of your steps are model calls — classify an inbound contract, research an account, draft a candidate reply — Sim’s canvas, per-run token traces and knowledge-base blocks fit the work more directly.

Pricing reality

Cloud Free is $0 with 1,000 one-time credits and one workspace. Pro is $25 per user per month with 6,000 credits a month plus a 2,000-credit weekly refresh. Max is $100 per user per month with 25,000 credits a month plus 4,000 a week, and adds sandboxes and Slack Connect. SSO, self-hosting support and dedicated support are Enterprise only, at custom pricing.

One credit is $0.005. Every run costs 1 credit as a base charge. Model calls on Sim’s hosted keys are billed at provider list price times 1.1, converted to credits. Bring your own key and the 10% markup disappears — you pay the provider directly and only the base charge stays. Overage is off by default. Turn on on-demand billing and extra credits cost $0.005, invoiced every time unbilled usage reaches $100.

Work one flow through. An agent run whose model call costs $0.02 at list price is 4.4 model credits plus 1 base, so 5.4 credits. Pro’s 6,000 monthly credits cover about 1,100 of those runs. With your own key the same allowance covers 6,000 runs, and the $120 in tokens goes to your provider. The difference between the two is the 10% markup, not a multiple. The model bill is the real cost either way.

Self-hosting the core is $0 in licence fees. The documented floor for 1-5 users is 2 CPU cores and 12 GB of RAM, and 4 cores with 16 GB for 5-50 users. Our estimate for a 16 GB VM with managed PostgreSQL on a mainstream cloud is $80-130 a month, which is roughly twice the footprint of Activepieces.

Best for

The ops engineer at a 20-200 person company building agent-heavy workflows who needs the code inside its own perimeter, or who needs to ship those workflows to customers. The scoped case where Sim wins outright: a legal-ops or CS team running a document pipeline — ingest contracts or tickets into a knowledge base, classify with an agent block, write results to a table, expose the workflow as an MCP tool the team calls from Claude — self-hosted on its own keys.

Not for

Teams whose automations are mostly SaaS-to-SaaS plumbing with no model in the loop. Zapier or n8n do that with deeper, older connectors. Also not for a buyer who needs a vendor with balance-sheet scale: a $7M Series A is small next to n8n’s $180M Series C and SAP’s 2026 secondary at a $5.2B valuation.

Versus the alternatives

• n8n — the category leader for technical teams, from $23/month on cloud. Pick it when your use is internal so the licence costs you nothing, and you want the largest template library, community and hiring pool.
• Zapier — the largest installed base, from $19.99/month. Pick it when flows are short, nobody wants to run infrastructure, and connector breadth beats agent depth.
• Gumloop — the fastest-growing AI-native entrant, with a $50M Series B led by Benchmark in March 2026; Pro is $37/month. Pick it for non-engineers running model-heavy pipelines who will never self-host.
• Activepieces — the other permissive-licence option, MIT core, from $16/month. Pick it over Sim when the work is integration-heavy rather than agent-heavy, or when a 12 GB RAM floor is too much for the box you have.

If none fit and your team writes code, build the agent in a code-first framework such as LangGraph or the Claude Agent SDK and run it on the scheduler you already operate. Flowise, which Workday bought in August 2025, announced its shutdown in summer 2026 — a reminder that a visual layer is the part you can lose.

Watch-outs

• Apache-2.0 covers the core, not the enterprise features. SSO, SCIM, audit logs, access control, data retention, session policy and white-labelling live in apps/sim/ee/ under the Sim Enterprise License, which allows evaluation and non-production use only. Guard: list the controls your security review mandates before the pilot. If SSO or audit logs are on it, price Enterprise then, not after the workflows are in production.
• Mothership and the Copilot need a sim.ai key when self-hosted. The chat module requires a COPILOT_API_KEY issued by sim.ai, so the natural-language layer calls Sim’s cloud. Guard: if data must stay inside your perimeter, disable the chat module and build on the canvas, with agent blocks pointed at your own provider keys or Ollama.
• It is pre-1.0 and ships fast. Several releases a week is good for features and bad for a production box that auto-updates. Guard: pin the Docker image to a specific version tag, upgrade on a staging instance first, and export deployed workflows before each upgrade.
• “1,000+ integrations” is not 1,000 native connectors. The December 2025 HN launch cited 138 integrations and the README now claims 1,000+, without saying how the count is defined. Guard: before committing a flow, confirm the exact trigger and actions you need exist as native blocks for your CRM or ATS, and budget time to build an MCP or HTTP call where they don’t.