mark_l_watson

That would be my question also. I like it when companies have easy to sign up for, pay as you go models. Being able to buy $5 worth of tokens and get an API key - in less than a few minutes - is ideal.

What you are suggesting might sound difficult to some people, it is possible: in the last week I co-wrote (with Antigravity with Claude as the backend) an Emacs package for agentic coding that also just uses ‘rg’ for finding relevant code for the context, call out to a model, and handle creating a diff with merging tools. I love using my own code that provides inside Emacs vibe coding and I would suggest others try building the same thing.

EDIT: here is a link of what I wrote just for my own use: https://github.com/mark-watson/coding-agent

Pi does seem special, at least to me. The only close to happy experience I have had coding with a small local model has been with Pi. Here is how I ran Pi and I was impressed how well it worked with a 9B model:

OLLAMA_CONTEXT_LENGTH=16384 ollama launch pi --model qwen3.5:9b

To get close to a happy experience with Claude Code or OpenCode with small local models I had to use 28B to 32B local models.

I can’t wait to try it. I set up a new system this morning with OpenClaw and GLM-5, and I like GLM-5 as the backend for Claude Code. Excellent results.

I was thinking the same thing: better to run models using a local service not in the wen browser. I use Ollama and LM Studio, switching between which service I have running depending on what I am working on. It should be straight forward to convert this open source project to use a different back end.

That said this looks like a cool project. It is so valuable writing projects like this that use local models, both for tool building and self education. I am writing my own “Emacs native” agentic coding harness and I am learning a lot.

I am lazy: when an LLM messes up parenthesis when working with any Lisp language I just quickly fix the mismatch myself rather than try to fix the tooling.

I agree with you, but a Python REPL in Emacs (using the ancient Python Emacs support) is very nice: initially load code from a buffer, then just reload functions as you edit them. I find it to be a nice dev experience: quick and easy edit/run cycles.

Same experience. I like Haskell a lot but I am not great at Haskell programming. LLM based coding agents are useful for helping with runtime errors, library versions, etc. (and as other people here have said, for tedious stuff like cleaning up Emacs customizations, etc.)

Thanks for the Scheme setup examples. I have created very simple skills markdown files for Common Lisp and Hylang/hy (Clojure-like lisp on top of Python). I need to spend more effort on my skills files though.

Hey, thanks for mentioning Datalevin (https://github.com/datalevin/datalevin) - I hadn’t seen that before.

This looks good. I had recently removed all Apple dev tools from my Macs - decided I didn’t want to spend my time any longer supporting a platform that requires submitting personal use apps to the Apple Store just to make them persistent (not time out).

For Ruckus, I will reinstall Xcode, etc.

Other people have mentioned LispPad that is very good.

Interesting, and not quite my experience. While I do get better agentic coding results for Python projects, I also get good results working with Common Lisp projects. I do have a habit of opening an Emacs buffer and writing a huge prompt with documentation details, sometimes sample code in other languages or if I am hitting APIs I add a working CURL example. For Common Lisp my initial prompts are often huge, but I find thinking about a problem and prompt creation to be fun.

The article mentions a REPL skill. I don’t do that: letting model+tools run sbcl is sufficient.

Oh, https://opencode.ai/zen looks good. I like pay as you go plans since I usually don’t use many tokens compared to vibe coders.

I regret paying Google for a one year AI subscription last spring (although it was a deep discount over the regular $20/month cost) because it has kept me from experimenting with many venders (but it was a fantastic deal financially).

I just put a reminder on my calendar to try OpenCode zen when my subscription ends.

I am not the person you are responding to but I have tried both: using OpenRouter and also giving a Chinese company $5 on my credit card to buy tokens. If I know what model I want to experiment with, I much prefer to just pay $5 and have plenty of tokens to experiment. On a yearly basis, this is a very tiny expense for the benefits of getting plenty of tokens to experiment with.

I agree with you and the GP post, even though I am an LLM enthusiast.

My primary interest is using small edge models to perform specific engineering tasks. In this pursuit I do like to use gemini-cli or Antigravity with Claude a few times a week as coding assistants, but I am using relatively few tokens to do this.

I also waste a lot of time, but this is fun time: experimenting with open source coding agents with local models just to see what kinds of results I can get. This is mostly a waste of time, but I enjoy it.

My other favorite use pattern: once or twice a week I like to use the iOS Gemini app in voice mode, and once a month also use video input. I really like this, but it is not life changing.

Externalities matter: I never use frontier LLM-based AI without thinking of energy, data center, and environmental costs.

thanks! I never thought of using -p for using claude and gemini for one-shots and in shell scripts before. Nice.

I agree with your criticism. I should have simply said that I had good results with gemma 4 tool use, and agentic coding with gemma 4 didn’t yet work well for me.

I wasn’t very clear, sorry. By my ‘own research’ I meant spending 90 minutes experimenting with Gemma 4 models for tool use (good results!) and a half hour using with pi and OpenCode (I didn’t get good results, yet.)

I spent two hours doing my own research before asking for Gemini’s analysis, which reinforced my own opinion that the gemini models historically have not been trained and target for agentic coding use.

Have you tried using the new Gemma 4 models with agentic coding tools?If you do, you might end up agreeing with me.

The article has a few good tips for using Ollama. Perhaps it should note that the Gemma 4 models are not really trained for strong performance with coding agents like OpenCode, Claude Code, pi, etc. The Gemma 4 models are excellent for applications requiring tool use, data extraction to JSON, etc. I asked Gemini Pro about this earlier and Gemini Pro recommended qwen 3.5 models specifically for coding, and backed that up with interesting material on training. This makes sense, and is something that I do: use strong models to build effective applications using small efficient models.

For about two years I experimented with writing local apps using local LLMs, but I often had to blend in a commercial web search API to make my little experiments useful.

I pay $13/month for Proton’s Lumo+ private chat LLM that contains an excellent built-in web search tool. I use it for everything non-technical, even just simple searching for local businesses, etc.

As an enthusiastic reader of books like Privacy is Power and Surveillance Capitalism, it feels good to have a private tool that is ready at hand.

I have been using Apple’s built-in system LLM model for the last 7 or 8 months. I like the feature that if it needs to, it occasionally uses a more powerful secure private cloud model. I also write my own app to wrap it.

I love to see new Emacs Lisp projects, BUT: personally I prefer a simple ‘pure Emacs standard library’ experience as much as possible. I have been using Emacs over 40 years and this return to simplicity is a new thing for me.

I used to have a Xerox Lisp Machine in the 1980s and dreamed to have Emacs be the ‘catch all’ environment like a Lisp Machine. Now I mostly just use Emacs to edit code.

Looks good, but only Linux is supported. I like spinning up VPS’s and then discarding them when I am done. On macOS, something I haven/t tried yet but plan to: create a separate user account.

Off topic but earlier today I asked Gemini to read this article and advise how to do the same things for OpenCode. I am fascinated with trying to get good performance from small local models.

I think the value right now in LLM code assist tools is in small projects: small reusable libraries or proof of concept “I want this app even though almost no one else does” types of projects.

For libraries: still probably mostly useful for personal code bases, but for developers with enough experience to modularize development efforts even for personal or niche projects.

I am bothered by huge vibe coded projects, for example like OpenClaw, that have huge amounts of code that has not undergone serious review, refactoring, etc.

I am biased: fully support the main ideas from Center Human Technology, books like Privacy is Power and Surveillance Capitalism, etc.

This verdict is a good thing for society. I am not anti-big tech, but big tech must accept responsibility for externalities created by profit first and screw over the public business models.

For me, this is a confusing/complex set of issues: I want to see humanity thrive and tech can play a positive role in that, but with the current ‘almost fascist’ collusion of our current regime with most big tech leaders, I am worried that we won’t get the kind of protective laws we need.

I appreciate this verdict!

I was enthusiastic about early TensorFlow in Swift efforts, sorry when the effort ended. My interest then flowed into early Mojo development for a while.

I wrote an eBook on Swift several ago but rarely update that book anymore. Count me as one of the many developers who for a while thought Swift would take over the world. At least Swift is a fun language to use, and now with LLM coding tools writing macOS/iOS/iPadOS apps is fairly easy.

I think this is a good learning project, based in a long perusal of the github repo. One suggestion: don’t call the CLI component of the project ‘claude’ - that seems like asking for legal takedown problems.