Sim - Open-source workspace for AI agents and workflows

Sim is an open-source workspace to build agentic workflows. Connect your AI agents and workflows to 1,000+ integrations and LLMs.

Add a comment

Replies

Best

I like that you're treating AI agents less like individual tools and more like a coordinated system. As companies adopt more agents, I think orchestration will matter far more than adding another capable model. The value shifts from what a single agent can do to how reliably they work together.

 exactly. Being able to orchestrate many agents to handle a single task easily will become more important over time. This concept already exists in ML, and it's called ensemble learning. Many "weak" models on average perform better than one stronger model.

 That's an interesting comparison. One difference, though, is that ensemble learning optimizes prediction quality, while companies ultimately care about decision quality. As more agents become involved, the challenge shifts from producing more outputs to determining which outputs have actually earned enough evidence to act on. That feels like the harder infrastructure problem over time

 I'd argue that they are more similar than different, but I agree that context is the most important aspect of the decision making process.

 That's exactly where I think the interesting question is.

Context determines which outputs become evidence, and evidence determines which agent gets trusted next. Once multiple agents start influencing each other, orchestration becomes less about execution and more about governing confidence.

I'm curious whether Sim eventually evolves more toward that layer than workflow automation itself.

Love the emphasis on replacing unnecessary LLM calls with deterministic code instead of assuming everything needs an agent. Just Curious have you noticed users converging toward a small set of reusable agent patterns over time, or is every team's workflow still highly bespoke? Congrats on the launch! 🚀

 

A pattern we're seeing often is when teams put a set of deterministic nodes first (fetching data, cleaning it, etc.) and then put the LLM-based nodes at the end to handle the last-mile and other nuances.

Quick question: when an agent workflow fails in production, how do you actually debug it? Like, is there a trace/replay view, or are you digging through logs across the different integrations it touched? Anyways, congrats on the launch!!

 thank you! Great question. You can see the full trace span then launch a debug agent on Sim that finds the issue with your workflow, solves the problem, then re-deploys your agent.

Congrats Emir!
Most agent builders just throw LLM calls at everything and the bill quietly becomes a problem at scale. How does Sim decide which steps get deterministic code versus an LLM call, is that the builder's call or does Sim suggest it

 great question! Sim will suggest using deterministic processes for steps that don't need to be brought into an agent.

When chatting with Sim, it will build everything for you, but the way we think about it is if that step requires some nuance then you should bring an agent in.

 That makes sense , using agents only where nuance is actually needed keeps the workflow lean and the costs predictable. Curious whether Sim tracks cost per workflow run over time so builders can see exactly which steps are eating the most tokens and optimize from there.

there's something called a "Sim" block so you can set up a workflow that triggers on every run and then keep track of the cost in tables or elsewhere.

How does the pricing work for an open-source project, especially once you start scaling beyond a handful of agents and hitting those 1,000+ integrations?

 it's free forever! Open source Apache 2.0.

Played around with it for an hour and honestly the drag-and-drop agent builder feels really snappy, hooked up a Slack-to-Notion workflow in like ten minutes without writing code.

 great to hear! Feel free to use the chat agent. It's really powerful and can get you set up in <60s.

i like the focus on production ready AI agents instead of simple demos. how does Sim monitor agent health and recover from failed workflow executions?


 great question. In logs, you're able to see the full trace span from a workflow run. If you error in that run, there's a debug from logs where the Sim agent can identify what went wrong and triage/fix the problem.

Congratulations

 thank you!

Open source is the reason I'd actually build a workflow on this instead of renting one that vanishes in a year. When a step in the middle of a run fails, does Sim let me resume from that step or does the whole thing start over? Re-running everything from the top is what made me drop the last couple orchestration tools I tried.

 great question. You can resume from a point in time in a workflow. If your workflow fails at point C, instead of re-running from point A, you can resume from C onwards.

connected my OpenAI agent to Slack and Notion in like 10 minutes, the drag and drop flow editor feels really natural

 feel free to try out the Sim agent builder! You can build in natural language in <1 minute.