g0t4

Officially only Azure OpenAI Services are supported for Windows Terminal Chat. In this guide I show how to create a service, deploy a model and configure terminal chat to use it.

I hate writing AppleScript code but I love what the Accessibility Framework affords to automate all sorts of mac applications. Script Debugger takes the pain out of finding and controlling UI elements.

OOB Win Terminal Chat only works with Azure OpenAI Services... so I hacked it to insert my own dotnet API to forward requests to any OpenAI compliant completions endpoint, and I demo how to set it up with ollama + codellama locally. Given the amount of time it takes to setup an Azure OpenAI Service and Deploy a Model... especially waiting a day or two for approval to use it... this is probably a faster way to get things up and running! So, why not!

After cajoling an Azure OpenAI Service (Endpoint) and deploying a model, Terminal Chat was up and running so I made a quick video overview of how to use it. Keep an eye out for a video soon about how to setup the Azure OpenAI endpoint.

Whipping up workflows (GitHub Actions) is much easier when using the official VSCode extension. Complete static schema and dyanmic data. Links to repos and docs for actions. Static "compile time" like checks. All makes life much easier than memorizing and/or relying on copilot alone. BTW, if you want to learn more about GitHub Actions, I am working on a new Pluralsight Course that goes into depth about it, keep an eye out for that in the next month or so.

My first impressions taking copilot workspace for a test drive. The thumbnail should say it all :)

Don't mess with scrolling. And stop it with playjacking a video.

It's time for people to grow up and solve real problems.

My first impression using Windows Recall... everyone needs to take a chill pill, it's basically the Snipping Tool with an Image Viewer. Here's a separate video w/ my setup: https://youtu.be/pY5-bzfd6QM

Setup Windows Recall using the AmperageKit repo instructions on a Parallels VM on my M1 mac. And then enabling snapshots. Here's my Recall app test drive video and opinions: https://youtu.be/uDmMFw6KvOU

I use multiple accounts on the same machine that both have access to the same set of streamdecks. Synchronizing profile changes was a hassle via backup/restore. So I did some digging and found how to sync the profiles via dropbox. This would work between multiple machines too.

What if we had a shell / language to reliably write english "code" and run it instead of needing to know arcane languages and commands?

I hate activating venvs, so I wrote some code to automatically detect and activate venvs via the current directory in fish (and zsh) shell. And then deactivate when I leave the context of a given venv. It triggers on a change in pwd so I can keep using all my favorite tools to CD around the filesystem (i.e. z) and it just works! https://github.com/g0t4/dotfiles/blob/9d726b25415307a5566120ef97e982512ccbf010/fish/load_last/auto-venv-on-cd-specific.fish https://github.com/g0t4/dotfiles/blob/9d726b25415307a5566120ef97e982512ccbf010/zsh/universals/3-last/auto-venv-on-cd-specific.zsh#L8

Quick overview of DuckDuckGo's new AI Chat. A tour of the models available. Plus I was scolded by Claude for asking who is POTUS... such an amazing model!

Time for another round of euphemisms to do things the way they've always been done: wrong.

Nothing like driving key business decisions by measuring lines of code and suggestion acceptance rates...?

The ag (silver searcher) command is a cross platform way to quickly find files by content and/or filename. And then simply command click to open the file in vscode (thanks to iTerm2's semantic history). https://github.com/ggreer/the_silver_searcher

nektos/act allows local testing of workflows (GitHub Actions) so you don't have to push every change to test it. Refine your steps locally and then push to GitHub Actions only after you know they're working.

I reimplemented my command line copilot as an iTerm Script. Primarily to stream response chunks as they arrive and to easily use the helper when SSHing into servers.

A walkthrough of the script I use to control how Semantic History opens files in command output by cmd+clicking on a file and then dispatching the open request based on the file type and current directory.

iTerm 3.5.0 introduced a ton of new features, including a new Command Info View that shows information about each command that has been run. Just cmd+click to find timing and exit code w/o re-running commands. Or copy the command, output or directory. Or edit the command in the composer. All just a few clicks away. Definitely worth knowing about, especially when troubleshooting.

On the heels of 3.5.0, 3.5.1beta2 has a few noteworthy changes, including using a local model with the AI features!

Apparently this model isn't censored, and has a sense of humor. Btw, https://www.youtube.com/watch?v=vnkLCoJwsk0 here is the previous video when I introduced pyautogen that generates code and executes it.

An agent model to automate interactions between models, in this case to have an assistant that generates code and a user proxy execute that code.

Plugging groq.com into my CLI copilot. Fastest API and its free, includes access to llama3-70b