Openclaw works like a buggy early AI

[wintry schooner]

i tried out openclaw, and i set out with two particular tasks just to start out: track a package i had coming in and let me know when the status changes, and occasionally check up on the health of my ceph cluster and notify me if it was unhealthy.

also this was on an LXC i dedicated just for openclaw

the best it was able to achieve, i asked the one main agent it came with in plain english to do those things. with some back and forth, we worked out a way to get playwright downloaded and installed on that lxc and a site it could query to get the tracking info, and because its environment wasnt part of the ceph environment, i gave it a proxmox read only api key to use to get info for the ceph cluster.

both of those tasks, it had worked out a way to at least get me the info for them after figuring it out. and i checked that info, the details the AI told me matched what i saw from individually going check on these things myself.

after those methods of getting the info was worked out, it had also scheduled some cron tasks, both the system cron and the openclaw cron. but they werent working quite right. the system cron would poll the endpoints it had set up and put them in a log file in the openclaw workspace. but there was nothing that checked on those logs regularly. or the openclaw cron would be instructed to check on the ceph cluster health, but whatever chat that spawned with those instructions, the AI inside would respond along the lines with "sure i can do that, but id need to do so and so, want me to do that now?", and thered be nothing to respond to that, and every time that cron would run, itd be that same type of thing, maybe just worded differently.

(continued)

[rancid plume]

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

[wintry schooner]

(continuation)

when i realized the issue, in some new chats, i tried manually asking it how the status of those things were. and it was almost like starting over to set up those things from scratch all over again. it was trying to use ceph in its own environment, it was telling me it cant access a package tracking without an api, that type of stuff, same basic steps it was going through when we initially worked it out the first time. i reminded it, it had already figured out and was able to do these things before. it searched its memories, usually not finding the right memories, recollecting when we were still in the middle of working this stuff out. at one point it even ended up recreating a similar separate solution to the one that was already worked out. other points it was hung up that whatever particular issue it was able to recollect up to was just some issue that was preventing this from working, like that was the conclusion, when it wasn't. it was a mess.

i have yet to get it to go out of its way to send me a chat about anything at any point. only time i heard from it is purely from it directly replying to my queries. the only time it technically reached out to me first, it wasn't even the AI talking to me. it was one of the ceph health check scripts it had set up to directly send a message to me with the API response of its health without going through any AI first.

even if it did start some background research type of thing, which i struggled to get it to do in the background instead of tying up the main chat thread with the research, itd usually never respond to me the results of that research. i had to ask it how it went.

(continued)