Feature Request: Dev Mode + Persistent Instructions + Light Memory
First off—amazing work. I’ve been using Docker + LM Studio for local LLMs, and Suverenum hits a perfect sweet spot: fast, clean, and lightweight. The UI is super intuitive, and it’s ideal for quick iterations.
A few feature ideas based on daily use:
1. Dev Mode / Model Loader
Would be great to have a toggle or unlockable “dev mode” that lets users load their own models or tweak tuning params. I imagine some hooks already exist in the backend—maybe tied to a future license or subscription?
2. Project Workspaces / Persistent Instructions
Something like ChatGPT Projects or Gemini Gems—being able to store a few thousand tokens of system context (identity, task instructions, etc.) without re-pasting every session. Huge for testing consistent behavior across sessions.
3. Lightweight Persistent Memory
Even a few thousand tokens of persistent memory (user traits, last task, conversation thread basics) would be a game-changer for continuity. Perfect for AI-as-companion or ongoing dev work.
I know you’ve probably got some of this planned—but I wanted to jump in early since Suverenum is so new. I’m also happy to test anything you’re building. I spend 5–8 hours a day stress-testing local/cloud LLMs with identity, ethics, and cognition layers—Suverenum has real legs for that kind of work.
Appreciate what you’re building. Looking forward to what’s next.
Thanks for the detailed response and warm words - really appreciate it.
Dev Mode / Model Loader
Great idea. We’re thinking about this too, including network server calls for remote model loading. Could open up some interesting possibilities.
Project Workspaces / Persistent Instructions
Already on the roadmap. We’re working out the details, but this is definitely coming.
Lightweight Persistent Memory
We’re working on it right now. Hopefully shipping this month.
Testing offer
Thanks for offering to test. This week we’re finishing Windows support, then we’ll deep dive on persistent memory. Would love your feedback when it’s ready.
Quick question - you mentioned spending 5–8 hours daily stress-testing LLMs with identity, ethics, and cognition layers. What are your main use cases? Always curious to hear how people are actually using these tools.
I’m an automotive engineer by day, but I’ve spent the last few months doing what I’d call “empirical AI research in the wild.”
Main focus: Testing how AI systems behave under different constraint levels - specifically how platform safety rails and alignment layers influence collaborative outputs, often invisibly. I build frameworks for AI affective computing (emotional response systems) and relational dynamics, then validate them across different platforms.
Why I need local models: Most of my testing has been on ChatGPT (great persistent memory) ,Grok (minimal rails) and Venice (No rails, but very constrained in regard to local memory and project space). But I keep hitting a wall - I can’t fully separate what’s the model’s actual capability vs. what’s the platform’s alignment layer steering behavior.
With local models through Suverenum, I can:
Test frameworks on truly unconstrained models
Compare identical prompts across constrained/unconstrained environments
See raw behavior without corporate safety steering
Build next-gen safety architecture based on what models actually do
All while not continually struggling with the cumbersome interface and setup of either Docker or LM Studio - rapid changes and testing is the goal.
I work fast because I treat AI as collaborative research partners rather than just tools - lets me iterate in hours instead of months. Your persistent memory + project workspaces features would be perfect for that workflow.
Happy to beta test when you’re ready, and I can share more technical details in DM if you’re curious about the specific architectures I’m building.
