Sim is an open-source workspace to build agentic workflows. Connect your AI agents and workflows to 1,000+ integrations and LLMs.
Replies
Best
the launch story really resonates because many builders have experienced tool sprawl. what was the hardest engineering challenge in bringing workflows, memory and integrations into one platform?
@olivia_bennett7 the most difficult part was scaling our workflow loads to be able to handle multi-hour or multi-day runs. Now, you can run workflows for days on end!
Report
Congrats on the launch! The shared workspace concept makes a lot of sense. can agents securely share context while maintaining role based permissions?
@barnaby_lloyd great question! You can create permission groups that lock down agent capabilities and give scoped access.
Report
replacing expensive LLM calls with deterministic logic is a smart design choice. do users get visibility into where tokens are being spent inside each workflow?
Congrats on the launch @emirkarabeg The token efficiency angle is so underrated. Finally someone is doing the math before burning the budget. What are the most popular integrations so far?
@kate_ramakaieva thank you! The most popular used to be Supabase before we added in our Sim tables integration. Now, it's searching/scraping tools like Exa and Firecrawl.
Report
Congrats on the launch! The open source angle is what caught my eye, most agent workflow tools lock you into their cloud. Curious about one thing, when an agent fails mid-workflow, does the whole chain retry or can it resume from where it stopped? Wishing you a great launch day!
@henry_s_jung great question! You can actually run from a point in the workflow based on data from the previous run. If A -> B -> C and you failed on B last run, you can re-use the data from A to re-run just B and C.
Report
Open-source makes this especially interesting. What advantages have you seen from teams self-hosting Sim versus using a hosted AI workspace?
@amjad_shaik self-hosting allows you to have complete data residency. Using Sim cloud allows teams to move fast initially. There are certainly benefits to both approaches.
Report
@emirkarabeg Thanks for clarifying! That trade-off between speed and data residency makes sense. Have you seen most teams eventually migrate to self-hosting, or do many stay on Sim Cloud long-term?
1,000+ integrations is impressive!! I'm curious, how do you think about long-tail services? Do you expect to build and maintain every connector yourselves, or is there a strategy for letting the ecosystem extend Sim over time?
@luki_notlowkey great question! We have an agent on Sim that writes the initial integration based on public docs from the tool we're integrating with. Then we have a suite of tests and we check over it briefly ourselves. It allows us to ship any integration in <5 minutes.
The shared memory across agents in the same workspace is cool 👍 can you also set boundaries so certain agents only access certain data, or is it all-or-nothing?
@ding_hao that's right! You can guard agents by only giving them access to certain operations (I.E. Gmail draft instead of Gmail write).
Report
@emirkarabeg The per-node flexibility between chat, canvas, and code is what other tools get wrong. Curious, which mode do most of your users end up living in?
Replies
the launch story really resonates because many builders have experienced tool sprawl. what was the hardest engineering challenge in bringing workflows, memory and integrations into one platform?
Sim
@olivia_bennett7 the most difficult part was scaling our workflow loads to be able to handle multi-hour or multi-day runs. Now, you can run workflows for days on end!
Congrats on the launch!
The shared workspace concept makes a lot of sense. can agents securely share context while maintaining role based permissions?
Sim
@barnaby_lloyd great question! You can create permission groups that lock down agent capabilities and give scoped access.
replacing expensive LLM calls with deterministic logic is a smart design choice. do users get visibility into where tokens are being spent inside each workflow?
Sim
@emily_carter18 inside your logs, you're able to see every trace span and token cost.
FuseBase
Congrats on the launch @emirkarabeg The token efficiency angle is so underrated. Finally someone is doing the math before burning the budget. What are the most popular integrations so far?
Sim
@kate_ramakaieva thank you! The most popular used to be Supabase before we added in our Sim tables integration. Now, it's searching/scraping tools like Exa and Firecrawl.
Congrats on the launch! The open source angle is what caught my eye, most agent workflow tools lock you into their cloud. Curious about one thing, when an agent fails mid-workflow, does the whole chain retry or can it resume from where it stopped? Wishing you a great launch day!
Sim
@henry_s_jung great question! You can actually run from a point in the workflow based on data from the previous run. If A -> B -> C and you failed on B last run, you can re-use the data from A to re-run just B and C.
Open-source makes this especially interesting. What advantages have you seen from teams self-hosting Sim versus using a hosted AI workspace?
Sim
@amjad_shaik self-hosting allows you to have complete data residency. Using Sim cloud allows teams to move fast initially. There are certainly benefits to both approaches.
@emirkarabeg Thanks for clarifying! That trade-off between speed and data residency makes sense. Have you seen most teams eventually migrate to self-hosting, or do many stay on Sim Cloud long-term?
Sim
@amjad_shaik teams that start on Sim Cloud end up staying long-term. It's more common for teams that start locally to move to the cloud.
AgentKey
Sim
@luki_notlowkey great question! We have an agent on Sim that writes the initial integration based on public docs from the tool we're integrating with. Then we have a suite of tests and we check over it briefly ourselves. It allows us to ship any integration in <5 minutes.
Besimple AI
this is super cool!
Sim
@yi_zhong3 thank you!
The shared memory across agents in the same workspace is cool 👍 can you also set boundaries so certain agents only access certain data, or is it all-or-nothing?
Sim
@ding_hao that's right! You can guard agents by only giving them access to certain operations (I.E. Gmail draft instead of Gmail write).
@emirkarabeg The per-node flexibility between chat, canvas, and code is what other tools get wrong. Curious, which mode do most of your users end up living in?
Sim
@abhiranjan_mehta most users work inside chat end-to-end.