LiteBin has multi-node support but a simpler deployment flow. From my local machine, I build a custom Docker Compose image with profile customizations for Hermes, compress it, and send it directly to the VPS for deployment — no Docker image registry needed. This simplifies things, but it has its own set of problems. Our custom image was a 900MB tar file that needs to go from my machine to the VPS. The current image upload doesn't support resumability. If it stops mid-way, I have to start over from scratch. I added a retry to the LiteBin CLI today — earlier, after 3 failed uploads it would exit and I'd have to recreate the image too. Now at least it retries the upload without rebuilding. These experiments are helping me find the gaps in LiteBin and see the real use cases as well.
Photo by Alexander Sinn
whoami
codinglog
Field notes, experiments, and the occasional rant. A log of what I'm learning, breaking, and building — plus the random stuff that doesn't fit anywhere else.
tools
The web framework this site is built on. Content-first, zero JS by default.
Headless CMS driving the dashboard. Open API over a SQLite file.
One compose file, the whole stack runs anywhere.
Fast, disk-efficient package manager. Replaced npm for me.
latest_posts
--> alllatest_notes
--> allSpun up an Oracle Ampere VPS — 2 OCPU, 12GB RAM, Mumbai region — for the Hermes Docker setup. Decided to use LiteBin (l8bin.com) as the deployment tool. Instead of running it standalone, I made this new VPS an agent node to my existing LiteBin master in Europe (Netcup VPS). Didn't work. Jio (my ISP) has connectivity issues with European servers and the connection keeps timing out. Couldn't deploy the Hermes setup. Will retry tomorrow. There's also a design issue with LiteBin I hit today. Right now, the agent node can only communicate through the master. So when I deploy an image, the tar file has to travel from my machine to Europe first, then from Europe to Mumbai. That's a lot of unnecessary latency. I could fix this by making LiteBin talk to the agent directly during deployments. But that needs more planning. For now, I'll retry the deployment tomorrow and deal with the architecture later.
Spent the last day setting up a Docker Compose for Hermes Agent with: Custom onboarding flow Skills pre-configured to set Neural Watt as default LLM provider Now I can configure this for my brother — he gets his own profile. Onboarding is triggered from Telegram, asks questions, and helps set up an Obsidian vault based on his needs. Next plan: deploy this on free Oracle tier and see how it goes.
Had to fight with flows in Directus. The agent was struggling and failed then I helped it, but finally we succeeded.
$ leave_a_message
Spotted something interesting, or want to point me at a tool? Drop it here.