For someone who's already deep into LangGraph or CrewAI, is there an import path, or would they basically be starting from scratch rebuilding their agents in Sim?
The shared context/workspace part is what I'd stress-test first. In agent workflows I've seen the brittle bit is less the prompt and more schema or credential drift across Slack/Notion/Supabase. Curious how you debug/version a workflow when one of the 1,000+ integrations changes?
Report
Kind of interesting that 90%+ of usage moved to chat when the product started out canvas-only. Usually builder tools go the other way, people start with prompts then graduate to manual/visual control once they need precision. Makes me think their user base skews more solo-builder/ops than engineer.
With 1,000+ integrations wired in, how predictable is token spend once an agent starts chaining calls across three or four of them in one workflow?
Report
As workflows grow over time, keeping everything organized usually become hard than building the first version. Wht helped your team keep large workflows easy to understand and maintain as more agents, integrations and triggers were added?
Report
Emir, congrats on the launch. That live integrations counter ticking past 1000 is such a nice touch, makes the scale feel real. Upvoted!
Report
the "deterministic steps wherever it counts" line caught me because that's the exact thing I ended up hand-coding after my n8n token bill got stupid. how does Sim decide what becomes real code vs an LLM call, is that automatic when you build by chatting or do I have to go mark steps myself? because if the chat builder defaults everything to LLM calls the cost problem just comes back quietly
Report
Congrats on the launch Curious how you are handling routing when a message could plausibly belong to two categories, does the AI pick one or surface it in both?
Replies
Swytchcode
This is really amazing. So do you have templates for the workflows?
How is it different to N8N or Zapier?
Sim
@chilarai no, most teams just build in the chat! No need for templates.
See: https://www.sim.ai/comparison
What's on the roadmap that you're most excited about but can't talk about yet? 👀
Sim
@chen_hao3
Two letters. U. I.
For someone who's already deep into LangGraph or CrewAI, is there an import path, or would they basically be starting from scratch rebuilding their agents in Sim?
Sim
@diego_joaquin1 that's a great question. You can pass your code/repo to the Sim agent and it will build it from end-to-end.
Context.dev
huge congrats!!
Sim
@yahia_bakour3 thank you Yahia!
The shared context/workspace part is what I'd stress-test first. In agent workflows I've seen the brittle bit is less the prompt and more schema or credential drift across Slack/Notion/Supabase. Curious how you debug/version a workflow when one of the 1,000+ integrations changes?
Kind of interesting that 90%+ of usage moved to chat when the product started out canvas-only. Usually builder tools go the other way, people start with prompts then graduate to manual/visual control once they need precision. Makes me think their user base skews more solo-builder/ops than engineer.
With 1,000+ integrations wired in, how predictable is token spend once an agent starts chaining calls across three or four of them in one workflow?
As workflows grow over time, keeping everything organized usually become hard than building the first version. Wht helped your team keep large workflows easy to understand and maintain as more agents, integrations and triggers were added?
Emir, congrats on the launch. That live integrations counter ticking past 1000 is such a nice touch, makes the scale feel real. Upvoted!
the "deterministic steps wherever it counts" line caught me because that's the exact thing I ended up hand-coding after my n8n token bill got stupid. how does Sim decide what becomes real code vs an LLM call, is that automatic when you build by chatting or do I have to go mark steps myself? because if the chat builder defaults everything to LLM calls the cost problem just comes back quietly
Congrats on the launch Curious how you are handling routing when a message could plausibly belong to two categories, does the AI pick one or surface it in both?