Agent 37 Cloud
Give every customer their own Hermes or OpenClaw agent
931 followers
Give every customer their own Hermes or OpenClaw agent
931 followers
Agent 37 is managed hosting for persistent agents like Hermes, OpenClaw and ClaudeCode. So you don't need to run them on Mac minis or VPS yourself. One API call gives each of your customers their own always-on agent, from $3.44/mo. Founders use it to ship vertical agents to their own clients without babysitting servers.
Agent 37
@vishnukool this is actually super clean
The “no babysitting servers, just spin up an agent per customer with one API call” is exactly the kind of abstraction people want but rarely get this smooth in practice
One thing I’m curious about: when you scale to lots of customer-specific agents, what becomes the hardest part to manage in reality? cost control, tool permissions, latency, or keeping each agent properly isolated without context bleeding between tenants?
Also love the idea of making vertical agents feel like products instead of infra pain
Excited to see what people build on this 🔥
Agent 37
@md_khayruzzaman Cost control is at the top of one of the things we need to manage and just the health of the servers in terms of RAM and CPU.
Thanks!
This solves the exact tradeoff I keep running into. I looked at self-hosting part of my own AI agent stack to save on a per-minute provider fee, and the math always won on paper but lost in practice: the real cost wasn't the server, it was becoming the person who gets paged when an agent goes down for a paying client. $3.44/mo always-on, not-my-problem-to-babysit is the right price for that peace of mind. Curious how you handle noisy-neighbor isolation between customer instances at that price point.
Agent 37
@david_marko That's great to hear. Would love to hear what your product us doing where you're seeing this trade off ? Here's my calendar if you wanna chat - https://cal.com/vishnukool/30min
We started off with a direct to consumer Hermes/Openclaw hosting and saw the same problems there where we run a large fleet of sandboxes. Interestingly what we've also noticed is when you have a large enough host and that coupled with the fact that agentic use cases like hermes, openclaw, claude code etc. are 90% idle CPU times, the noisy neighbor issue is rarely a problem.
what happens when one customer's agent goes rogue, infinite loops, runaway API calls to other services, anything that spikes cost or behaves unexpectedly? since each customer gets their own always-on instance, curious whether there's any per-agent resource capping or kill-switch a founder can configure, or if that's left entirely up to whatever the founder builds on top of Agent 37
Agent 37
@ansari_adin We do have resource monitoring to track and catch such things, but largely what a user runs inside their sandbox is upto them
@vishnukool monitoring helps you detect it, but curious if there's anything automated that acts on it, like a default spend cap or auto-pause if a sandbox blows past expected usage, or is detection mainly there so you (Agent 37) can see what's happening, while the actual response (pause, kill, alert the founder) is something the founder has to build themselves on top
Agent 37
@ansari_adin Oh yes we do have monthly and total spend caps you can set when creating an instance. In Addition there's also monitoring dashboard where you can see peak CPU usage, RAM usage, how many times RAM hit peak, CPU wait time etc. We're adding more every week, but these are some things that already exists
This is a sharp wedge for vertical-agent teams. The part I’d want to understand is the customer boundary: when each customer gets a persistent agent with tools, are credentials, approvals, memory, and action logs isolated per customer by default, or does the founder wire that layer around the hosted agent?
Agent 37
@blah_mad So the default templates we offer comes with LLMs pre-wired, 1000+ tool integrations like Gmail, Calendar, Notion etc. via Composio pre-wired and a few other things.
But for custom use cases you can also create your own custom Docker images that doesn't have these pre-wired if you'd want to integrate with your own APIs.
Got it. Custom images for teams that want their own API layer makes sense.
The piece I’d still want explicit is the tenant boundary: are Composio connections, memory, approval state, and action logs isolated per customer by Agent 37, or does the founder model that around each hosted agent?
Congrats. I wonder how does Agent 37 ensure complete data isolation between tenants sharing the same underlying infrastructure?
Agent 37
Thanks @crystalmei
We follow a similar architecture to Modal as described here - https://modal.com/docs/guide/security. Each tenant is containerized and virtualized using gVisor which keeps them all isolated. Do you have anything in mind that you're thinking of to try us out ?
The per-customer always-on agent from one API call is the part I'd build on — shipping a vertical agent without standing up a server per client is exactly the friction I keep hitting. The thing I'd test first: where does each agent's memory/state actually live, and is it isolated per customer and exportable if a client churns or I later want to migrate them off Agent 37? I'd want that state to be portable, not locked to the hosting.
Agent 37
@noctis06 Yes, it's isolated per customer and exportable as needed. You essentially have full control over the sandbox container and can do any operation within it, like move data out/migrate etc.
Full container control plus move-data-out is exactly what makes this safe to build on — thanks. Two things I'd still want to pin down: when a customer's agent restarts or you redeploy, does the sandbox state (installed tools, memory, configured integrations) persist by default, or do I re-provision it each time? And is there a clean export/import so I can snapshot one customer's setup and stamp it onto a new tenant instead of wiring each from scratch?
Agent 37
@noctis06 yes the customer's data is processed by default. Basically everything in the home folder is preserved by default, which is where most harnesses like open cloud Hermes or even Claude Code save data.
Since we allow Docker images, you can basically templatize any setup that you have in mind and then install and spawn instances to your customers based on that image template
StartupBase
The per-customer agent model makes a lot of sense for vertical SaaS. What I keep thinking about is observability: how does a founder show their client what the agent actually did?
Logs the end customer can understand, or at least something you can point to when they ask, "what did it do this week?" Is that on the roadmap or already there?
Agent 37
@attacomsian We do have observability at the container level and logs around that, but what happens inside a container is upto the user for now. We do provide default ready to use templates for hermes, openclaw etc. But it's extendible so that you can add-on observability as you see fit. Especially the one's I think you're referring to which is more for the end customer.