This looks really cool for DIYing workflows, especially since you seem to have a very useful selection of tools!
Did you build your own agent engine? Why not LangGraph?
Say I was building a general agentic chat app with LangGraph in the backend (as it seems to provide a lot of infrastructure for highly reliable and interactive agents, all the way up to a protocol usable by UIs, plus a decent ecosystem, making it very easily extensible). Could I integrate with this for DIY workflows in a high quality fashion (high-precision updates and control)?
Is there a case for switching out LangGraph‘s backend with Sim (can you build agents of the same quality and complexity - I’m thinking coding agent)? Could it interact with LangGraph agents in a high quality way so you can tap that ecosystem?
Can I use Sim workflows with my current agent, say, via MCP?
1. we wanted to have full control over the agent orchestration and the execution since we didn't like the abstractions that many of the existing frameworks had built, and didn't want to have dependencies in places we didn't need them. so, we built the orchestration and execution engine from scratch, allowing us to do neat things like human in the loop, settings that run the same block 10 times concurrently, etc.
2. this would kind of serve as a drop-in replacement for langgraph. you could build a workflow with an agent and some tools, perhaps some form of memory. then, just deploy that as an API, call it from your frontend, and consume the streamed response on your chat client and without the need to maintain any infra at all.
3. we have a generic code block and an api block used to call APIs for integrations that we may not have, and you can use those to plug (langgraph) agents into the Sim ecosystem.
4. we are adding in the ability to deploy your workflow as an MCP server in the next week, stay tuned :) in the meantime, you can deploy the workflow as an API and have the agent call it as a tool. moreover, you can use the workflow block in sim to call other agents/worklows as well, so its easy to encapsulate a lot of complexity in a `parent` workflow that you call that dynamically routes and uses different tools based on the task at hand
Their deployment stuff has been turning me off lately; everyone is rushing to monetize - which I understand and support - but I feel like Langsmith is creeping further and further into Langchain|graph and it makes me hesitant to invest. It’s giving AWS-like gentle but firm lock-in vibes, I wonder if they have any PMs from there.
I do like the way they’ve been able to leverage Langgraph workflows to build agents - it seems like the right abstraction to me - and I also feel their middleware approach is very Django-y which I also like. Are you enjoying their stack?
I’m only in the research phase of my hypothetical project so far, so I’m going more off of vibes than personal experience for now.
I’m interested in LangGraph because it seems the closest to an industry standard - every use case seems to be addressed with a tutorial (both first and third party) and there’s an ecosystem of already available graphs/agents. I’m aiming for both high extensibility (new use cases should be easily implementable) and high reliability. The LangGraph docs do a pretty good job at convincing me that they got the latter pretty nailed down. It seems like a hard enough problem to question a new solution on this.
I want to build a (highly reliable & controllable) UI for agents more than I want to build the agents themselves, so my hope is that LangGraph has the biggest ecosystem I can plug into.
They do have some funky lock-in attempts, for instance the LangGraph CLI, which acts as a server for their agent protocol (https://github.com/langchain-ai/agent-protocol), is proprietary. However (and this is what I consider indicative of a strong ecosystem) there’s a free reimplementation named Aegra: https://www.aegra.dev/
So here is a case that I wanted to implement in n8n a few years ago and it required quite heavy JS blocks:
- I want to check some input - pick one of your 138 blocks
- I want to extract a list of items from that input
- I want to check which items did I encounter before <- that's the key bit
- Do something for the items that have not been encountered before; bonus point for detecting updated and deleted items
- Rinse and repeat
It could be a row added to a CSV file, a new file dropped into a Nextcloud folder, a list of issues pulled from a repo, or an RSS feed (Yahoo! Pipes, what a sweet memory).
How good is the support for such a case in Sim? And did it get better in n8n?
this is actually a perfect use case, mostly deterministic workflows that need LLMs to fill in the gaps or do the knowledge work. As you mentioned, you can either add it as a row in a CSV file (sheets), use the baked-in memory block and treat it as simple storage, store the row in supabase, or use the knowledgebase. Basically, there are a ton of ways that this can be done that don't require you to maintain the memory solution yourself. you can even detect the updated and deleted items by keeping some sort of version-controlled snapshot of each row in the csv and updating it as you go.
I can't tell you whether it got better in n8n, but I can definitively say that this sounds like a great candidate workflow to build in sim :)
the agents would be great in the instructions where we need to `do something`, but asides from that is sounds like a pure orchestration task.
now, handling an integration to something like google sheets myself for a task this small is a nightmare, not to mention the separate table I'd need to keep to store the access token & refresh token and the permissions I'd need to get from google. on top of that, hosting it somewhere and then monitoring it.
Hey - could be a good use case for https://github.com/autokitteh/autokitteh - which gives you durable workflows over python. Since your logic is deterministic it's a simple python script that stores the history in memory, and autokitteh will take care of the persistancy aspect.
Excited to try this out, I've been looking at LangFlow and similar tools for doing DAG workflows. Sure, I could prompt or try to do an MCP or a claude skill for my utility workflows, but they aren't strongly followed and I want to, where possible, make each AI agent call be smaller, like a function.
This is definitely going to be given a try tomorrow morning. I think first up will be something easy and personal like going through the collection of NPC character sheets in my recent campaign and ensuring all NPCs have the following sections with some content in them, and if not, flagging them for my review.
its a conservative estimate, but primarily because there's a socket server that runs alongside the main container, so you never have to manually save because changes are debounced, broadcasted to any other users on the canvas, and sent to the socket server which persists it to the DB
I've been pronouncing it as "nanites" in my head since the first time I've seen it. Dunno why and now I feel just a tiny silly, though I still prefer my pronunciation.
Needless to say I never had to actually say "n8n".
n8n uses a "Sustainable Use License"—source available, but not OSI-approved open source. This means you can only use it for internal business purposes or non-commercial use. Sim is Apache 2.0.
Edit: I checked out your materials, I see. Copilot; smart - one could probably do this themselves with the self hosted, and those that would do this aren’t good customers for you. Those that don’t want to bother will use the free tier and possibly convert, and maybe you’ll get some enterprise out of the bargain. Good luck!
What is your business strategy, if someone like AWS could treat you like Elastic or Redis? Premium addons? As a consumer I have no problem with n8ns license, it’s their technology that I think could be done better by a fresh competitor.
Copilot and a few modules are sim.ai-only, but self-hosters can access them with an API key. The core—execution engine, all integrations, deployment—is fully Apache 2.0.
We aimed for feature parity with n8n and improved much of the workflow building experience along the way. We also added what we think agentic workflows needed (mentioned above). Would love your feedback.
Hi Waleed, it's great to see your commitment to improving workflow management with Sim. It sounds like you've faced some significant challenges with existing tools, and I'd love to hear more about your experience. What specific integrations have you found most challenging to work with?
Node-RED is great for IoT/edge/data flows. Sim is built specifically for AI agents—native LLM support, tool-use control, structured outputs, token-level observability, etc.
afaik, there are actually a ton of unique IoT integrations that node-red has. a majority of their nodes/flows have accompanying physical devices and sensors
You actually can, if you self-host there are environment variables to control what models are available to the copilot but it’s tuned to Azure for the time being. We can work on generalizing it further and documenting it better
Azure is just fine, as long as it's documented someplace. I'll take a look, although I also couldn't find prebuilt Docker images referenced in the compose.local file (I will look into what is being built into ghcr.io)
the prebuilt images are in compose.prod file only, not in compose.local.
since the copilot is a managed service, you’d be setting those azure credentials in your .env and the copilot would call into your azure openai deployment.
1) Is there something that n8n does that's not possible on Sim?
2) Is there a node to create custom stuff like n8n's code node? Possibly nodejs and python
3) Is it easy to import packages from npm or pip
1) functionally, not that I'm aware of. the only thing right now is running from a specific block onwards, something that we are rolling out this week. you'll be able to run the workflow in debug mode and get really granular information about each node, the inputs and outputs, and resume from any point in the workflow with mock data
2) there is, we also have a code node that uses E2B to run code in isolated sandboxes. it supports python and ts/js
3) yes, in the code node you can use frameworks/libs since we do RCE
for loops we use two sentinel nodes with a backwards edge, and before each iteration, we check the condition and update loop variables.
sentinel -> body -> sentinel (condition with backwards edge to first sentinel)
in the UI, this is just represented as another block, and depending on the varying types of loops you can either define a collection or the number of iterations
at the moment, we don't support 'loops in loops' on the client-side, but not for any other reason asides from it becoming confusing for users. since we don't actually make copies for each loop, it wouldn't be a performance issue.
We keep all workflows running on Sim cloud backwards compatible always. The idea is that you build, deploy, and never have to make modifications again unless you want to.
If we release a breaking change that requires a migration for existing local workflows, we release notice at least a few weeks ahead of time/bake it into the db migrations.
Incase there are significant changes made, everything is versioned so you opt-in to upgrading.
This looks really cool for DIYing workflows, especially since you seem to have a very useful selection of tools!
Did you build your own agent engine? Why not LangGraph?
Say I was building a general agentic chat app with LangGraph in the backend (as it seems to provide a lot of infrastructure for highly reliable and interactive agents, all the way up to a protocol usable by UIs, plus a decent ecosystem, making it very easily extensible). Could I integrate with this for DIY workflows in a high quality fashion (high-precision updates and control)?
Is there a case for switching out LangGraph‘s backend with Sim (can you build agents of the same quality and complexity - I’m thinking coding agent)? Could it interact with LangGraph agents in a high quality way so you can tap that ecosystem?
Can I use Sim workflows with my current agent, say, via MCP?
1. we wanted to have full control over the agent orchestration and the execution since we didn't like the abstractions that many of the existing frameworks had built, and didn't want to have dependencies in places we didn't need them. so, we built the orchestration and execution engine from scratch, allowing us to do neat things like human in the loop, settings that run the same block 10 times concurrently, etc.
2. this would kind of serve as a drop-in replacement for langgraph. you could build a workflow with an agent and some tools, perhaps some form of memory. then, just deploy that as an API, call it from your frontend, and consume the streamed response on your chat client and without the need to maintain any infra at all.
3. we have a generic code block and an api block used to call APIs for integrations that we may not have, and you can use those to plug (langgraph) agents into the Sim ecosystem.
4. we are adding in the ability to deploy your workflow as an MCP server in the next week, stay tuned :) in the meantime, you can deploy the workflow as an API and have the agent call it as a tool. moreover, you can use the workflow block in sim to call other agents/worklows as well, so its easy to encapsulate a lot of complexity in a `parent` workflow that you call that dynamically routes and uses different tools based on the task at hand
Their deployment stuff has been turning me off lately; everyone is rushing to monetize - which I understand and support - but I feel like Langsmith is creeping further and further into Langchain|graph and it makes me hesitant to invest. It’s giving AWS-like gentle but firm lock-in vibes, I wonder if they have any PMs from there.
I do like the way they’ve been able to leverage Langgraph workflows to build agents - it seems like the right abstraction to me - and I also feel their middleware approach is very Django-y which I also like. Are you enjoying their stack?
I’m only in the research phase of my hypothetical project so far, so I’m going more off of vibes than personal experience for now.
I’m interested in LangGraph because it seems the closest to an industry standard - every use case seems to be addressed with a tutorial (both first and third party) and there’s an ecosystem of already available graphs/agents. I’m aiming for both high extensibility (new use cases should be easily implementable) and high reliability. The LangGraph docs do a pretty good job at convincing me that they got the latter pretty nailed down. It seems like a hard enough problem to question a new solution on this.
I want to build a (highly reliable & controllable) UI for agents more than I want to build the agents themselves, so my hope is that LangGraph has the biggest ecosystem I can plug into.
They do have some funky lock-in attempts, for instance the LangGraph CLI, which acts as a server for their agent protocol (https://github.com/langchain-ai/agent-protocol), is proprietary. However (and this is what I consider indicative of a strong ecosystem) there’s a free reimplementation named Aegra: https://www.aegra.dev/
So here is a case that I wanted to implement in n8n a few years ago and it required quite heavy JS blocks:
- I want to check some input - pick one of your 138 blocks
- I want to extract a list of items from that input
- I want to check which items did I encounter before <- that's the key bit
- Do something for the items that have not been encountered before; bonus point for detecting updated and deleted items
- Rinse and repeat
It could be a row added to a CSV file, a new file dropped into a Nextcloud folder, a list of issues pulled from a repo, or an RSS feed (Yahoo! Pipes, what a sweet memory).
How good is the support for such a case in Sim? And did it get better in n8n?
this is actually a perfect use case, mostly deterministic workflows that need LLMs to fill in the gaps or do the knowledge work. As you mentioned, you can either add it as a row in a CSV file (sheets), use the baked-in memory block and treat it as simple storage, store the row in supabase, or use the knowledgebase. Basically, there are a ton of ways that this can be done that don't require you to maintain the memory solution yourself. you can even detect the updated and deleted items by keeping some sort of version-controlled snapshot of each row in the csv and updating it as you go.
I can't tell you whether it got better in n8n, but I can definitively say that this sounds like a great candidate workflow to build in sim :)
I really like windmill.dev which should support your scenario just fine
N8n can definitely do this.
They recently added native tables, albeit still just a few data types, you can store stuff in and use in workflows.
But does that require AI agents? Well, maybe the extraction step, if it's not CSV but a general-case web page.
Maybe in the middle, processing the items - classifying, summarizing.
But the post bills the tool as an n8n alternative. Therefore, I am evaluating it as such. Solid basics before the AI whizbang.
the agents would be great in the instructions where we need to `do something`, but asides from that is sounds like a pure orchestration task.
now, handling an integration to something like google sheets myself for a task this small is a nightmare, not to mention the separate table I'd need to keep to store the access token & refresh token and the permissions I'd need to get from google. on top of that, hosting it somewhere and then monitoring it.
Hey - could be a good use case for https://github.com/autokitteh/autokitteh - which gives you durable workflows over python. Since your logic is deterministic it's a simple python script that stores the history in memory, and autokitteh will take care of the persistancy aspect.
Excited to try this out, I've been looking at LangFlow and similar tools for doing DAG workflows. Sure, I could prompt or try to do an MCP or a claude skill for my utility workflows, but they aren't strongly followed and I want to, where possible, make each AI agent call be smaller, like a function.
This is definitely going to be given a try tomorrow morning. I think first up will be something easy and personal like going through the collection of NPC character sheets in my recent campaign and ensuring all NPCs have the following sections with some content in them, and if not, flagging them for my review.
sounds super cool! let me know how it goes
> You can run Sim locally using Docker, with no execution limits or other restrictions.
This is big. Thank you.
hope you enjoy :)
Looks interesting. The 12GB minimum RAM requirements seem quite steep though. Why so much?
its a conservative estimate, but primarily because there's a socket server that runs alongside the main container, so you never have to manually save because changes are debounced, broadcasted to any other users on the canvas, and sent to the socket server which persists it to the DB
What does “n8n” stand for? I’m assuming it’s a shortening of a longer word, like k8s.
I've been pronouncing it as "nanites" in my head since the first time I've seen it. Dunno why and now I feel just a tiny silly, though I still prefer my pronunciation.
Needless to say I never had to actually say "n8n".
i believe it stands for “nodemation”
Correct, check the "Our name" section at: https://n8n.io/press/
naaathaaan
Is that intoned like Ricardo Montalbán's "Khaaaan!" ?
I think more "Fenton!"...
https://youtu.be/3GRSbr0EYYU
I wonder how the free open-source in this stacks up to the free open-source in n8n.
n8n uses a "Sustainable Use License"—source available, but not OSI-approved open source. This means you can only use it for internal business purposes or non-commercial use. Sim is Apache 2.0.
Edit: I checked out your materials, I see. Copilot; smart - one could probably do this themselves with the self hosted, and those that would do this aren’t good customers for you. Those that don’t want to bother will use the free tier and possibly convert, and maybe you’ll get some enterprise out of the bargain. Good luck!
What is your business strategy, if someone like AWS could treat you like Elastic or Redis? Premium addons? As a consumer I have no problem with n8ns license, it’s their technology that I think could be done better by a fresh competitor.
Copilot and a few modules are sim.ai-only, but self-hosters can access them with an API key. The core—execution engine, all integrations, deployment—is fully Apache 2.0.
We aimed for feature parity with n8n and improved much of the workflow building experience along the way. We also added what we think agentic workflows needed (mentioned above). Would love your feedback.
Hi Waleed, it's great to see your commitment to improving workflow management with Sim. It sounds like you've faced some significant challenges with existing tools, and I'd love to hear more about your experience. What specific integrations have you found most challenging to work with?
A bit of feedback, the readme gifs are a little too fast, it's hard to tell what exactly is happening.
thanks for the feedback, I was actually thinking this the other day. we'll slow them down.
Also on the topic of UX. Dark mode on main site to prevent the blinding?
coming soon
How does this compare to say, Node Red?
Node-Red is the original.
N8n and its clones are attempts to copy and commercialize it with a closed-source license.
Node-RED is great for IoT/edge/data flows. Sim is built specifically for AI agents—native LLM support, tool-use control, structured outputs, token-level observability, etc.
There is nothing in Node-Red that is specific to IoT.
afaik, there are actually a ton of unique IoT integrations that node-red has. a majority of their nodes/flows have accompanying physical devices and sensors
No, you're misremembering.
The only weird integration that Node-Red has is MQTT, but that isn't really anything IoT-specific, just a niche message broker.
This is the other major project in this space, although there are many, so finding the differences can be challenging.
https://www.activepieces.com | https://github.com/activepieces/activepieces
This looks really cool, but not being able to use my own LLM endpoints for the Copilot is an instant turn-off.
You actually can, if you self-host there are environment variables to control what models are available to the copilot but it’s tuned to Azure for the time being. We can work on generalizing it further and documenting it better
Azure is just fine, as long as it's documented someplace. I'll take a look, although I also couldn't find prebuilt Docker images referenced in the compose.local file (I will look into what is being built into ghcr.io)
the prebuilt images are in compose.prod file only, not in compose.local.
since the copilot is a managed service, you’d be setting those azure credentials in your .env and the copilot would call into your azure openai deployment.
Few qs:
1) Is there something that n8n does that's not possible on Sim? 2) Is there a node to create custom stuff like n8n's code node? Possibly nodejs and python 3) Is it easy to import packages from npm or pip
1) functionally, not that I'm aware of. the only thing right now is running from a specific block onwards, something that we are rolling out this week. you'll be able to run the workflow in debug mode and get really granular information about each node, the inputs and outputs, and resume from any point in the workflow with mock data
2) there is, we also have a code node that uses E2B to run code in isolated sandboxes. it supports python and ts/js
3) yes, in the code node you can use frameworks/libs since we do RCE
How does it deal with loops? I’ve often see workflow builders struggle at that?
for loops we use two sentinel nodes with a backwards edge, and before each iteration, we check the condition and update loop variables.
in the UI, this is just represented as another block, and depending on the varying types of loops you can either define a collection or the number of iterationsand specifically nested loops. if you're spinning up full runtime copies for each loop, you're gonna have a hard time
at the moment, we don't support 'loops in loops' on the client-side, but not for any other reason asides from it becoming confusing for users. since we don't actually make copies for each loop, it wouldn't be a performance issue.
Development seems pretty rapid, how often are breaking changes forcing workflow modifications to keep updated with the latest versions?
We keep all workflows running on Sim cloud backwards compatible always. The idea is that you build, deploy, and never have to make modifications again unless you want to.
If we release a breaking change that requires a migration for existing local workflows, we release notice at least a few weeks ahead of time/bake it into the db migrations.
Incase there are significant changes made, everything is versioned so you opt-in to upgrading.
Thanks, and that sounds great. On the backend what are you using for the DAG stuff to make it durable? Temporal?
We actually wrote our own serializer & execution engine and for long-running jobs, we defer to trigger.dev
for the workflow DAG, what type of backend are you guys using? is it like temporal or self-built durable workflows
we built an execution engine ourselves https://github.com/simstudioai/sim/tree/main/apps/sim/execut... and for the infra for background jobs, we use https://trigger.dev/
Open Sourced until we get rug pulled..
Everything that’s available today in our repo will be Apache 2.0 forever, so you don’t have to worry about that :)
This is amazing! Thank you all for building this!
thank you! hope you enjoy
[dead]