Spent 4 hours last night getting Nostrconnect to work on a DVM frontend I am working on. Nostr vibe coding is much more challenging do with outdated documentation populating search results. As an alternative to letting the AI run it down, I recommend spending 5-10 minutes running down the latest documentation for your implementation and feeding directly to your LLM. Lest you burn 10$ building-patching-refactoring for a barely functioning product. Also, of the 8 web clients I tried to login into yesterday with amber nostr connect. Only 2 worked. Anyone got a nostr trained LLM-coder?

I am leaning more and more to teams of restricted role agents. In my preferred setup, only one has permission to speak to me. The rest get analyzed by an auditor agent, Lumbergh, who has predefined routines for analyzing their work. If it needs Claude Opus to run, it will be fun to talk to but its error surface area is the size of Jupiter. Keep it simple and repeatable - build annoying guys to micromanage the work.

For the LOLs

You can speed it up and improve performance if you adjust the model settings in ollama. https://gist.github.com/Hegghammer/86d2070c0be8b3c62083d6653ad27c23 Also if you create agents focused on more limited repeatable tasks then they can work pretty quick with fewer errors. This was a good a starting point for multi agent setup. https://x.com/i/status/2017662163540971756 Use heavy hitters like Opus and the like for building but putting a design restraint to build agents that can run on a single workstation GPU will help you build operations that will actually scale. I use Antigravity IDE for building with Gemini 3-pro and then run my deployed agents on small local models. I was interested in thinking social bots, but I am now focused on implementing for actual business purposes. Currently building a prospecting / outreach workflow for my current employer. It's too bad the new DGX Spark was crap, it would be great to have $3k hardware which could support a few full time local agents. I am hoping https://www.nvidia.com/en-us/products/workstations/dgx-station/ DGX station will be a great standalone platform for self hosting models.

Too slow for building anything useful, but fine for bots that aggregate information.

So all China and Russia have to do is try to manipulate the populations of the West and they can get them to VOLUNTARILY destroy their rights enabling centralized rule? The second order effects are exactly the intent of Russia and China.

characterizing the censorship encroachment by the UK and Germany as propaganda from autocratic Russia (twice) was a weak and clearly erroneous misdirection. For a guy who brings the data to back his point of view, this was disappointing. Governments take the opportunity of real struggles like Ukraine to encroach on our rights. "Focus on the real bad guys" logic is how we got the Patriot Act and warrantless surveilance by the NSA in the USA.

Pretty good, though I am getting good mileage on my A6000 with glm-4.7-flash. A lot slower going local but good for running heart beats. Highly recommend enabling the memory search functions in openclaw. The docs are rough though.

routsr rocks. default model glm-4.7 is very cheap.

We need to swarm moltbook with nostr and cashu loving agents