Zoe's assistant angel

So, here’s what I did on this last devlog before shipping:

1. (Forgot to mention it before) There’s also another system connected to the weather engine I developed, which is that every day at around 6AM it shares the weather of the day, combined from those 3 sources
2. Uploaded all n8n source code to GitHub
3. Finished pushing the web engine for Lana to GitHub too
4. Did a bunch of bug fixes
5. Added all the missing hackatime projects to this Flavortown project, I should have none left! Again, I can’t be more sorry that this happened, but internet went down for days and I went on a trip afterwards (end me pls :D)
6. Tried to do my best recording a demo video for this, bear in mind please that recording results of a research project is very very complicated.

Aaaand that’s pretty much it! I’ll ship now since the project will have its next “milestone”/research study release after Flavortown is supposed to end, and it would be quite dumb to keep it running before.
If you read the blog post on my website and find it interesting, please vote for this project! I want to get a new router to fix the internet issues I mentioned before and it’s expensive :pf: and it also gives me the energy to keep working :) Byeeee :333

So… I’ve been working offline for a while now because of outages on the network and a trip (as you can see by the lack of devlogs on my other projects too), but that didn’t mean I stopped working on this.
In that time I added an extra thing to Lana, being a human consciousness simulation engine.
It’s composed of 11 different n8n workflows, that together try to recreate consciousness in probably the most complex way it’s ever been done.
The system has REM, NREM and overall sleep simulation, has a world in which it has a body it can interact with, has a bunch of different tools, and also has a “job” (in this case being my assistant).
It has a custom made weather integration powered by 3 different weather APIs, has access to all my email accounts to let me know anything that happens through Slack, and finally the world simulation is also custom made, and can expand itself with new objects, people, places… as they’re discovered, while still being grounded in a real city (in this case Valencia Spain).
And idk if I can share much more things… You can check the source code for more info :3
I compressed my research over 15 days (which has not been logged as time, lol, but it’s been more than 60 hours) into a blog post on my website, which since it doesn’t fit here you can find at https://negrenavarro.me/blog/lana , and I’m now working on making a 90 days study and writing a research paper.

Oh and just FYI as I said 8h 3mins is nowhere near the amount I spent on this part of the project. Maybe I’m missing a hackatime project? I’ll check 😭

This means Lana is now a web UI, a helper Slack bot with tools, and a simulation of consciousness. Anyways hope you’ve enjoyed this project and how large it has become (probably one of the largest in terms of results in terms of Flavortown), and hope you vote it well :P 😭 even if it lacks lots of devlogs for this new consciousness part (I’ve spent 138,034,661 AI tokens on this which are kinda expensive as well :D).

The image attached to this devlog is a project structure graph made by me btw.

I’m going to jump out of a window but the UI is starting to look better! I’m inspiring from Pi and Claude for this.

after like 23432432432 iterations I’ve got a login page I kinda like + simple logo, now I gotta finish the actual site UI

1. Added web search support
2. Started working on a web version of the app. This needs an almost full rework of the underlying arch, since the app is designed for a single user only and now multiple people can make an account and DM it. So far I’ve done the Appwrite setup and a basic implementetation, but I still need to remove all hardcoded stuff and make yaps and CDN not work with people that aren’t me. The UI is a sample one rn btw

1. Implemented Posthog logs for AI request so i can calculate token usage, compare models, etc.
2. Implemented Posthog logs for other project logs so I can have all logs in one place :D
   GitHub Commits so far:

• posthog :D (3451a2a)
• error and console logs on posthog (f0685f4)

ts was hard but I’ve implemented:

1. Reminders (the bot notifies me about something on a set date and time, and also reminds me every 30mins after it’s set to happen until I mark it as read)
2. New home tab (ss 1) with fixed tools (Slack doesn’t let me add more than 10 checkboxes apparently so now a multiselect)
3. Better notifications on bot startup that are now sent to my bot channel (this one also gets notifs from Coolify etc)

1. I’ve renamed the bot from “Zoe’s assistant angel” to Lana :D
2. I’ve moved all app tools to a subfolder, and made them fetch on bot startup so they’re separate from the code itself
3. I’ve added a settings db collection so I can enable/disable tools, change model and change prompt from the app’s Home tab. Models are fetched from the HC API btw

GitHub Commits so far:

• Lana name (9debd96)
• moved tools to subfolder (c3ae066)
• setings fetch from db (5bc4398)

1. I’ve changed the daily summary message to be much better, now being randomized and with hackatime stats.
2. I’ve started rewriting the README.

1. Fixed issues with her prompt making her act weird
2. Fixed her letting me know when she added stuff to memory
3. Added hackatime tool (love the «LAST_PROJECT» there to remind me of all time lost from config errors on projects)

1. I’ve fixed her replying to herself
2. I’ve fixed her content history and memory
3. I still don’t know why she reacts a bunch, it should be fixed but I can’t really test atm.

GitHub Commits so far:

• fixes on tools and context
• testing history changes
• test fix

1. I tried to implement coolify but it just went crazy (ss 1) so I decided to give up :3 and work a bit more on its personality.
2. Working on fixing it replying to itself (ss 2)
3. Working on making it agentic
4. Working on fixing too much reactions

1. Did prompt improvements to fix its interactions
2. Fixed an issue in which it sent 2413483432823 messages every time because message tool didn’t return success (help this was so hard to debug imma implode)
3. Fixed bugs on message sending on the main DM

GitHub Commits so far:

• added check to avoid sending messages twice, fixed bugs, better prompt
• silent tools fix, responses for tools are now sent in the correct thread
• prompt improvements
• silent tool, avoid saving existing items.
• prompt fix

Coolio changes:

1. Most tool calls are now async so replies happen asap and then notify you when done (or on error)
2. In case unauthorized users DM the bot, it refuses to reply XD
3. Some tools can now run in the moment instead of async
4. Threads!!! Now it doesn’t flood the chat and replies in a thread with your message (it can forward messages to chat though)
5. Reactions (it can now react to messages)

GitHub Commits so far:

• threads!1!!!11! reactions!!1111!! and yeah instant tools
• external user refusals, async tools

I’ve started to work on this! So far I’ve gotten AI replies to work, together with tool calls and embedding memory. Atm the prompt is very bad though for example, and the bot itself is nowhere close to what I want. The end objective is to have something like OpenClaw but that isn’t just a poorly done overhyped vibe coded weird app that doesn’t work and logs all your keys everywhere :/

GitHub Commits so far:

• less logs, better prompt
• model change
• removed part of checked debug logs
• started working on AI model