grantcarthew

People tend to forget about token usage when discussing languages with agents. Python is one of the best, but why use interpreted languages now. Also, the ecosystem is well...

Go is quite good from a token usage point of view. Rust is not so good due to the complexity of the language.

I see AI agent programming language choice as only two now: Go or Rust.

Go for most systems, Rust for high performance.

You're not considering one of the important factors, token usage. Rust is a complex language and burns through tokens compared to Go. Python is good with tokens, but why would you use an interpreted language when you don't write the code. Same with TypeScript. Compile it.

I see language choice as only two now: Go or Rust.

Go for most systems, Rust for high performance.

I built "start" so I didn't have to worry about any of this.

Using it I don't need skills, memory, subagents, a specific agent CLI. It defines roles, tasks, context out of the box.

I made it for me and my family though. I don't expect interest outside of that.

https://github.com/grantcarthew/start

I posted a Show HN post recently. It has not had a single glance of interest. I think it is dead now anyway. I've spent months on the project I posted, but no interest from anyone.

Yes, Atlassian's CLI support is shocking.

This is for Confluence. Markdown all the way:

https://github.com/grantcarthew/acon

Jira, same, markdown all the way:

https://github.com/grantcarthew/ajira

Other agent tools:

https://github.com/grantcarthew/homebrew-tap

There are a few around. I originally had a bash script https://github.com/grantcarthew/scripts/blob/main/lib/archiv..., this is a little more developed

Built for AI agents to consume web content efficiently.

This is a pattern I'm using with AI agents to increase context population speed for commonly used documentation.

I've used it to read authenticated pages with Chromium. It can be run as a headless browser and convert the HTML to markdown, but I generally open Chromium, authenticate to the system, then allow the CLI agent to interact with the page.

https://github.com/grantcarthew/scripts/blob/main/get-webpag...

I put an unsecured open FTP server on the internet about 20 years ago, just to see what would happen.

Within half a day I had some pirate "marking" his claim to my FTP server, then he/she started uploading a game. I deleted everything and left it open again.

It was a long time ago, so I don't remember all the details, but all the pirates would create directories inside directories, upload files, then mark it with their mark. All of this was scripted I gather.

After a while, I set up a file system watcher that deleted subdirectories. This gave me an FTP server I could use for anything. I shut it down a few months later.

Interesting though.

I recently automated this process, however, in a very different way.

- CLI run for every job you want to apply for (this is important)

- JavaScript (Deno) with Puppeteer to run the JS for the page

- Create a directory for all the artefacts <yyyy-mm-dd-ms-pagetitle>

- Save the webpage link (artefact)

- Take a screenshot of the page (artefact)

- Extract the HTML (artefact)

- Convert HTML to Markdown with a CLI (artefact)

- Send Markdown to the Grok API to extract just the Job Description as Markdown (artefact)

- Send Job Description and Autobiography to Grok API to generate a Resume (artefact)

- Send Job Description and Autobiography to Grok API to generate a Cover Letter (artefact)

- Use pandoc to convert the Markdown Resume and Cover Letter into Open Document Format (LibreOffice) (artefacts)

The important differences here are:

- You need to find the job you are interested in. Why automate this?

- Run the CLI `job-hunter https://job.site/jobid` (50sec runtime)

- Open the ODF documents, review, edit, save (human involved is important)

- Use a bash script running LibreOffice CLI to convert ODF documents to PDF

- Review the PDFs

- Manually click the apply button on the site and upload the documents

I also keep a spreadsheet with the details for each job I apply for so I can track interactions, think CRM for job applications and recruiters. This could be automated, however, I got a job so have lost interest.

Points of interest:

- Markdown is a fantastic format in general, but for LLMs as prompts and documents, it's awesome.

- If you just curl the page html, you don't get the recruiters email addresses in most cases, hence the use of Puppeteer.

- Having all the artefacts saved on disk is important for review before and after the application, including adding notes.

- By using an Autobiography that is extreme in detail, the LLM can add anything and everything about you to the documents.

- Use Grok and support Elon. OpenAI can stick their "Open" where it fits.

- I don't end up having to format the documents that are generated as ODF files, they look great.

I can apply for around 10 to 20 jobs in a day if I try hard. Most of the time it is around 5 because I am doing other things. They are only jobs I'm interested in though, and I can customise the documents. Also, If I am applying for a job that includes AI, I add a note at the bottom stating it has been generated by an LLM and customised.

There's probably more interesting points, but you get the idea.

My TODO list includes a CLI switch to only open the page in a Firefox profile so I can authenticate to the page. This removes the stupid "automate auth on ever job site" issue. Simply authenticate and keep the cookie in the hunter profile.

The repo is private for the time being, but I could make it public.

Edit: formatting.

I disagree, the landing page is just fantastic.

I like the product as well, however, I need to use it for a week to comment.

Nobody: Let's write all of our scripts in YAML

Me: !

haha, you guys.

Some haven't been updated for a while. Try the "software-evaluation.md" prompt, it's one of my best. I use it often. Use the bin/gp script to get them into your clipboard.

OK, you win.

No, it's related to Astigmatism. The eye lens gets distorted. Bright light shrinks the pupil, meaning the light goes through less of your lens and hence is not as distorted.

That's what the "Workspaces" are. Create a workspace, go into the settings then "Chat Settings" and configure the Prompt. There are more configurations available.

AnythingLLM supports Claude out of the box.

I'm using AnythingLLM for this now. https://anythingllm.com/

MIT License on GitHub

This week it got the X.ai API added which is what I'm using. Grok is not as good as Claude or ChatGPT yet, but don't bet against Elon.

https://x.com/ServeTheHome/status/1850917031421399543

https://x.com/elonmusk/status/1830650370336473253

It's the desktop solution I've been looking for. If you have a heap of prompts, you can create a Workspace for each. Super easy.

Here's my prompts if anyone is interested:

https://github.com/grantcarthew/notes/tree/main/Prompts

I only have two minor complaints with AnythingLLM: 1. No light mode (why does everyone want dark mode, it sucks for anyone older than 30) 2. It's electron

Edit: replaced cluster link with source (I think)

There are two killer features being no ads, and customise your search.

Don't want pinterest in your search results? Block it:

https://kagi.com/stats?stat=leaderboard

I discovered lsd because of this thread. It has a great tree output:

lsd --tree

Why is Anytype https://anytype.io/ not on this page somewhere? It should be.

I just whacked this together last night. I've been using it in another project but it was so useful that I wanted it as a standalone package.

I've been using Kagi since I could first get onto the product. As soon as it came out of Beta, I subscribed.

I had no idea how intrusive Google was. I love Google services. Google Cloud is far better than the others. That said, the search is just poison.

I now spend time looking at results, not searching for results.

I have to admit though, ChatGPT did make me consider dropping the subscription. Mr Chat has weakened the value of Kagi to me.

They have their own AI summarizer engine though: https://kagi.com/summarizer/index.html

That's a great suggestion @cosmaioan. I'm working on a Version 3 which will support SVG files. I might look in to adding this style of line. Thanks.

Thanks frantzme. I replied to your email however, for anyone else reading this, I don't have a Mac. That is a little of a handbrake.

Hi Omneity,

Isn't that just "Take Control"? Most video conferencing software packages have that feature. Zoom has an annotation feature.

I have some plans for it but none of them involve networking.

Hi Murple,

The Windows and Linux versions are the same code base. I originally had it as open source but decided not to keep it open.