Frannky

I know a person. Ex-Googly. Doing his startup. Spent one year on a crazy complex product. Investors do not get it. Users do not get it. He spends 99% of the time explaining why his ideas are so good. You ask to try the product, weeks pass and just slides and video demos. When you eventually try it, it's so confusing and nothing really works. I tried to make him understand that the constraint he needs to fit is users being able to understand rapidly why it's useful to them and that it should work. He does not care. He says it's about the story, and that the story will drive millions unlocking a super big team building his idea. I said that's cool but why in the meantime you don't just go with one thing that is useful and works and then procedurally evolve to your vision while you interact with the user and discover more about their problems. The answer was a one-hour speech on Google leveling system. Maybe he is right. Time will tell.

Thanks for reference the docs. For me an agent is an entity that you can ask something and it talks to you and try to do what you asked to do.

In this case if you have a server with an endpoint you can run opencode when the endpoint is called and pass it the prompt. Opencode then think, plan and act accordingly to you request, possibly using tools, skills, calling endpoints,etc.

No resources, I used Claude Code and did what I put in the original message. Experience easy thanks to Claude coding and deploying

I think you are mixing two concepts. I was just talking about having an LLM that is able to replicate human thinking, which is different then having a precise person's brains turned into LLM weights.

In that second case the problems you are saying emerge. But I can understand why you conflate the two, since having a model that works like a human may unlock the ability to dump the brain into model weights.

I model LLMs as searchers. Give the input search and they match an output. The massiveness of the parameters and training data let them map data in a way searching looks like human thinking. They can also permutate a little and still stay in a space that can overlap with reality.

The human brain may be doing a very similar thing though, search and permutation via searched rules. It may be doing it just in a functional way, with more ability to search on massive data that may be with holes but filled with synthetic data via mind subprocesses on learned rules.

I think machines can eventually get there, especially if we can figure out how to harness continuous models instead of discrete ones. And I have a feeling that functional analysis may be the key.

I don't use it for coding but as an agent backend. Maybe opencode was thought for coding mainly, but for me, it's incredibly good as an agent, especially when paired with skills, a fastapi server, and opencode go(minimax) is just so much intelligence at an incredibly cheap price. Plus, you can talk to it via channels if you use a claw.

Like this user has a comment history of hyping show HN

It may also be that I am just comment AI paranoid, but yeah, I find myself a lot guessing if there's a person behind a comment or not

I am very confused by the comments, they seem too excited for this... Are they real or paid bots? If they are real, kudos to OP

Can I run it?

I tried it once; it was incredibly verbose, generating an insane amount of files. I stopped using it because I was worried it would not be possible to rapidly, cheaply, and robustly update things as interaction with users generated new requirements.

The best way I have today is to start with a project requirements document and then ask for a step-by-step implementation plan, and then go do the thing at each step but only after I greenlight the strategy of the current step. I also specify minimal, modular, and functional stateless code.

The MCP server usually provides some functions you can run, possibly with some database interaction.

So when you run it, your codign agent is using AI to run that code (what to call, what parameters to pass, and so on). Via MCP, they don't pay any LLM cost; they just offer the code and the endpoint.

But this is usually messy for the coding agent since it fills up the context. While if you use skill + API, it's easier for the agent since there's no code in the context, just how to call the API and what to pass.

With something like this, you can then have very complex things happening in the endpoint without the agent worrying about context rot or being able to deal with that functionality.

But to have that difficult functionality, you also need to call an LLM inside the endpoint, which is problematic if the person offering the MCP service does not want to cover LLM costs.

So it does matter if it's an endpoint or an MCP because the agent is able to do more complex and robust stuff if it uses skill and HTTP.

I don't know. Skill+http endpoint feel way safer, powerful and robust. The problem is usually that the entity offering the endpoint, if the endpoint is ai powered, concur in LLM costs. While via mcp the coding agent is eating that cost, unless you are also the one running the API and so can use the coding plan endpoint to do the ai thing

Opus 4.6 is nuts. Everything I throw at it works. Frontend, backend, algorithms—it does not matter.

I start with a PRD, ask for a step-by-step plan, and just execute on each step at a time. Sometimes ideas are dumb, but checking and guiding step by step helps it ship working things in hours.

It was also the first AI I felt, "Damn, this thing is smarter than me."

The other crazy thing is that with today's tech, these things can be made to work at 1k tokens/sec with multiple agents working at the same time, each at that speed.

I have no idea how much the demand is for fine-tuned models. Is it big? Are people actively looking for endpoints for fine-tuned models? Why? Mostly out of curiosity, I personally never had the need.

What I want from an LLM is smart, super cheap, fast, and private. I wonder if we will ever get there. Like having a cheap Cerebras machine at home with oss 400B models on it.

I use another vps for FastAPI. I assume the vps with Zeroclaw will become compromised.

I don't use other people's skills. Just use Claude Code for adding endpoints and the relative skill.

Communication via Telegram bot, calendar via synced CalDAV server. Emails not connected; I don't need it personally.

LLM API is OpenCode Go Minimax. $10/month capped, I have never hit the limits.

I found ZeroClaw plus Hetzner to be a good option. I've been using it for a week, and it's stable and robust.

Complex abilities unlocked calling a FastAPI server with one skill for each endpoint

This is so cool! What are the cool projects in the space? I would love to start building and playing with these robots

I wonder what happens if I apply the same strategy to an automated shop. Claude code periodically proposes updates and automatically implements them, with revenue as the target function.I'll give it a try.

Cool idea! Does this work with electron apps? I tried automating some apps and the problem was that a lot of stuff was never reachable, only via screenshot + click

This is my experience too. Also, the ones not striving for simplicity and not architecting end up with giant monsters that are very unstable and very difficult to update or make robust. They usually then look for another engineer to solve their mess. Usually, the easy way for the new engineer is just to architect and then turbo-build with Claude Code. But they are stuck in sunk cost prison with their mess and can't let it go :(

I love it. I still use a 2020 M1 since I had not had a reason to change it. So smooth and fast and silent and thin. I don't really have a reason to buy a new one but if it broke or I lost it I would for sure buy this new one. Pros are too bulky for me. The only thing I miss I the ability to run local big models at high token rate.

Taking into account Tesla chips deliver way more intelligence on smaller hardware maybe we will get there, a glm5 model running locally at 100 tokens/sec would be really dope!

Damn I would love to buy it. In the past I tried different mods trying to get rid of google, the problem was always the same, lot of little annoyances making it very painful for daily usage. A de Googled phone without annoyances and security would be very cool.

Another interesting thing is that I haven't had any reason to buy a new phone in a very long time so we are probably in a time where the hardware is commodotized enough for motorola to be able to ship exactly what I need.

Never thought I would have think of routing for Motorola in 2026 but you never know!

I recently installed Zeroclaw instead of OpenClaw on a new VPS(It seems a little safer). It wasn’t as straightforward as OpenClaw, but it was easy to setup. I added skills that call endpoints and also cron jobs to trigger recurrent skills. The endpoints are hosted on a separate VPS running FastAPI (Hetzner, ~$12/month for two vps).

I’m assuming the claw might eventually be compromised. If that happens, the damage is limited: they could steal the GLM coding API key (which has a fixed monthly cost, so no risk of huge bills), spam the endpoints (which are rate-limited), or access a Telegram bot I use specifically for this project

I recently installed Zeroclaw instead of OpenClaw on a new VPS(It seems a little safer). It wasn’t as straightforward as OpenClaw, but it was easy to setup. I added skills that call endpoints and also cron jobs to trigger recurrent skills. The endpoints are hosted on a separate VPS running FastAPI (Hetzner, ~$12/month for two vps).

I’m assuming the claw might eventually be compromised. If that happens, the damage is limited: they could steal the GLM coding API key (which has a fixed monthly cost, so no risk of huge bills), spam the endpoints (which are rate-limited), or access a Telegram bot I use specifically for this project

I think what it means is that coding is difficult but predictable — in the sense that, you can solve it by throwing enough money at it. Figuring out what to build in a way that actually leads to product-market fit, on the other hand, is something you cannot solve just by throwing money at it. So in this frame, coding becomes 'the easy part' — not because it's truly easy, but because it can be solved relatively straightforwardly with resources

I think I'll wait until they are more reliable. For now, I use skills, but they just specify which endpoint to call. It should be also safer, different vps, no access to credentials but the bearer token.

Yeah, that's my experience too

both

I think unless you're doing simple tasks, skills are unreliable. For better reliability, I have the agent trigger APIs that handles the complex logic (and its own LLM calls) internally. Has anyone found a solid strategy for making complex 'skills' more dependable?