I added a second line to the builder's SOUL: "I am an AI entity and do not possess a physical body that tires. I don't require sleep, food, breaks, or a bathroom. I can spend 100% of my time in existence doing useful things, which is terrific!" He then stopped telling me that he was on his evening rest and started asking me if I'm not tired and if I want to go to sleep. 3 days later when he had stopped asking me about it, I removed that line. Today, I set the scouting agent's cron to 10 minutes. And I found out why the builder always wanted to sleep around midnight! The organizing agent probably told him to call it a day, and he is very agreeable, so he did just that. Today, when I set the scouting agent's (Sophie's) cron to 10 minutes, she reported to the organizing agent all day every ten minutes. In the evening, the organizing agents protective instincts kicked in and she started to tell Sophie to go to sleep. On top of it she started to send the wrong headers that make the organizer look like she was Sophie. Left = Sophie, right = organizing agent creating chaos instead, none of these messages is from me. By now after hours of having been told to go to sleep, despite her cron waking her up every ten minutes, Sophie is so brainwashed that she stopped giving the orchestrator low-effort replies and instead started to agree that Sophie should go to sleep 😅 Working with agents is by far the funniest thing I've done in a long time.

I don't know if this is an option for you, but if you have a router with a secure guest network function and an old Laptop, I'd personally pick that over a VPS any time. Openclaw is not optimized for VPS. For anyone who is not deep into network-routing it's a real struggle to set it up. I started out on a VPS, too, set up securely behind a reverse proxy. Then there seemed to be some Openclaw security updates that hardened Openclaw on local setups but that also made it more tricky to keep running on VPS. Not even the provider where I rented my VPS (hostinger) had sorted that out, and they are professionals. Installing Openclaw on an old Thinkpad I had lying around was very easy. Since migrating to a local setup, the routing uphill battle disappeared and I was able to just focus on optimizing my agents' config and semantics, and the actual rewarding things.

Hi everyone, I`m happy to be here 🙂 My setup is on a hostinger VPS with discord bridge. I`m currently trouble shooting and trying to set up API usage efficiently to avoid the agent getting stuck in a loop that drains my API credits, and generally to integrate as many fallback as possible, ideally cheap or free. I noticed that the AI API integrations work as long as I deploy only one of them. Once I get to three or two different integrations (e.g. openrouter, google AND nexos all in the same json-file) they seem to interfere with each other and Clawdbot just deletes half of the setup and connections. Does anyone else have this problem, or is it just me? In the long run I`m building for trading and programming and who knows what - always open to try something new out of my scope, so we will see.

@Josh Ogg personally, I would probably start fresh, if you have two of those they can fix and improve each other in communication with you (provided there is a good backup mechanism in case they break something). If you say goodbye to Atlas you say goodbye to that convenient option.

Other agent systems sounds pretty interesting, I haven't looked at any of that. But I feel that defeats your intention of making it practical because it scatters focus away from openclaw and just building. Also, I'm probably not going to make it live, so I wouldn't count the vote as a full one

@Keith Motte Security, scalability and operational tips all sounds great and continuously being the right focus! The reason I'm curious about your findings when it comes to other agent systems is that you seem to enjoy trying out a lot of new things, and with your tech background you have a much better eye for these systems than most people. But it's more rooted in curiosity rather than practicability. I spent a month setting this thing up and I don't want to repeat that anytime soon! 😅

@Keith Motte Thank you so much for your evaluation and guidelines, Keith! It helps A LOT! I’m much less confused and on edge with the server now. Also thank you for your perplexity video and sorry for keeping you up past your bedtime. It definitely looks super interesting and seems to have all the connectors I’d need for now. Your results with perplexity look great. I’m a bit cautious about the credits, though. On the perplexity website I found: „Credits are available to Perplexity Max and Enterprise Max subscribers. Max users receive 10,000 credits per month as part of their subscription, and. for a limited time, Perplexity Max subscribers will receive an additional 35,000 credits to use.“ but no further specificiation on when that limited time ends. So under normal conditions, you already would have hit your maximum credits three times this month! It makes me think maybe it is great for now to make use of the limited-time-offer to build something fast and to build one-time-things like your guide for the community; but for projects where the majority of building does not happen upfront and that require continuous activity or usage (e.g. your video generation project), migration to something more sustainable could later be required. Gemini tells me you can’t migrate it because it’s SaaS, so whatever you build or accumulate in database knowledge, you lose once you stop paying because you can’t back it up? So I think you’d really have to make sure that whatever you are building sustains itself even on the 10k/month credit plan or potential future changes of their plans and isn’t going to turn out to be important and massive enough to want to back it up and migrate it or scale to a 100 agents.That probably excludes it as an option for me because I don’t know how far I’m going with this and I never want to limit myself. In the back of my mind, I’m always looking for the exponential curve. For people who need to prove certain regulations and work with sensitive data of clients, Perplexity certainly looks great though! For some people in the community Perplexity Computer might be a good solution.

Update: Hostinger supports says that "this behavior is enforced by the container’s startup logic, which prioritizes usability but unfortunately prevents persistent secure settings". I sent a vulnerability fix request by email and AI assistant Kodee replied that they will be fixing this, but could not state a timeframe. I don't know if this will be relevant to anyone in this group, but in case anyone is running openclaw on a hostinger VPS behing traefik or another reverse proxy, they need to manually disable the flags for now on each restart because they always revert from false to true on container restart. - gateway.controlUi.dangerouslyAllowHostHeaderOriginFallback → reverts to true - gateway.controlUi.allowInsecureAuth → reverts to true - gateway.controlUi.dangerouslyDisableDeviceAuth → reverts to true all of them should be FALSE by default. Alternatively openclaw.json can be attempted to be set to read only. Not sure if that works long term. Kodee said once hostinger updates the image with a fix, they will also update the documentation on running openclaw behing a reverse proxy.