Jae Slackbot

V1.2.4-experimental | you can freely access JAE by sending a DM through the Hack Club workspace to the agent, or by joining the #-jae channel! <3

Jae can now access Stardance projects, and search the web with a configured SearXNG server! I also made sure that if you don’t configure them, the appropriate tools don’t get registered (to make your life easier)

Update is live on Slack, and the experimental version is available through the first pull request.

PR: https://github.com/JustAnEric/jae-slackbot/pull/1

Thanks for reading! (sorry for the slow updates, school has been on and I’ve also been reviewing projects for my ship payout)

README | Commit History

V1.2.3-experimental (haha, 1-2-3…) | you can freely access JAE by sending a DM through the Hack Club workspace to the agent, or by joining the #-jae channel! <3

After a lot of trial and error, I’m proud to say that Jae now has ONE tool that hasn’t been released yet, called the run_command tool.

On the running of this tool, an Alpine Linux container will be spun up for Jae to experiment inside! The packages I included in the Dockerfile (or image) for Jae to use are Rust / Cargo / Rustup, Python 3 and NodeJS.

Docker Compose manages the behind-the-scenes for the Jae Execution Server and Docker-Socket-Proxy (thanks, Tecnativa :D) and if you were to run the bot on a different machine from the execution server, you can use the handy environment variables declared in the bot EXECUTION_SERVER=<IP>:<PORT> and EXECUTION_SERVER_TOKEN=.... Of course I had to include a token, otherwise it just wouldn’t be that secure.

A next bit to this might be persistence for a single conversation thread, or setting resource limits so the bot doesn’t accidentally run a crypto miner. And, then even allowing Jae to interact with programs could be a great addition.

Update is live on Slack, but of course I will be monitoring + Jae will be a little hesitant to run commands.

Thanks for reading!

README | Commit History

V1.2.2-experimental | you can freely access JAE by sending a DM through the Hack Club workspace to the agent, or by joining the #-jae channel! <3

I have been working a major update, so basically I created an execution server for Jae the AI to use… and I don’t want to push it to main yet just because of safety issues and concerns I’ll still need to fix up.

The GitHub tree for these changes I made in 9 hours is here if anyone is interested: https://github.com/JustAnEric/jae-slackbot/tree/dev-execution-server

Expanding on my last devlog, I couldn’t brainstorm much other uses for tools other than this, and making these changes, Jae will probably be making a considerable turn in its primary capabilities (from not being able to generate code to being able to generate code).

I’ve also added in a Slack Canvas (this wasn’t said in any other devlog, so I’ve decided to link it here): https://hackclub.enterprise.slack.com/docs/T0266FRGM/F0BTEH85NKX

Thanks for reading!

README | Commit History

V1.2.1 | you can freely access JAE by sending a DM through the Hack Club workspace to the agent, or by joining the #-jae channel! <3

Jae now has feedback buttons! No settings or dashboard has been created yet for this functionality, but it is still going to be useful for gathering feedback from everyone into how Jae can improve.

I feel like all my images on the devlogs are so out of context, sorry about that… Main point is, we now have those buttons I guess.

README | Commit History

V1.2.0 | you can freely access JAE by sending a DM through the Hack Club workspace to the agent, or by joining the #-jae channel! <3

Even more cleanups, and I changed it all to TYPESCRIPT!!!
Also, now there’s tool calling included… I really need suggestions for tools to include, whatever you’d like to see… I beg you all to make me busy. ;)
I still want to work on making tools and thinking available outside of streaming mode, but I’m SUPER lazy and there’s no other APIs yet supported… just admit streaming is better at this point.
A TOOLS_ENABLED=true environment setting turns on this functionality, and tools can be found inside src/tools, I’ll create some clarity by editing the README soon for the creation of your own custom tools, I want to make it as easy as possible, of course.

README | Commit History

V1.1.1 | you can freely access JAE by sending a DM through the Hack Club workspace to the agent, or by joining the #-jae channel! <3
Some minor cleanups here and there, more refactoring and bug testing, also devised a new title generation feature that will be coming soon for threads as Slack allows that apparently… will probably get the AI to generate the title after 3 messages. There will be more updates, just been a little slow with it due to school. Hopefully I will get tool-calling working by the end of the week. Maybe even a change to TypeScript, as IntelliSense with VS Code on NodeJS is really bad.
README | Commit History

V1.1.0 | you can freely access JAE by sending a DM through the Hack Club workspace to the agent, or by joining the #-jae channel! <3
Today, I’ve now finalized the thinking aspect of the Slack bot! The thinking functionality currently only supports streaming mode (so I will work on getting the latter done in the near future, I only know how to show the tasks via progression.)
It uses Slack’s task_update and plan_update methods, tested on all Gemma 4 models. The Gemma 4 models shine for this purpose because they output a consistent, structured trace instead of just rambling for most of the time.
If the model chosen doesn’t support the format set in the code, the thinking trace won’t be displayed to the user and instead will display a Thinking… text.
This took way longer than expected, especially because I started on it late at night.. but at least it’s finally done.
It’s enabled by default in the GitHub repository’s new .env.example file, just as streaming mode is enabled by default.
README | Commit History

V.1.0.0
I’ve now gotten the main aspects down: the bot setup, model responses through an Ollama API, and you can access it on Slack through the #-jae channel or sending a DM. The future calls for tool-calling and thinking blocks. I’ll probably get the thinking blocks done in a few more hours, I’m learning about @slack/bolt as I go :-)
Commit History | README