LLMs Forget. Redprint Remembers
Instead of chasing bigger models and longer context windows, Redprint shifts the focus to compression, memory, and reasoning. It is designed to reduce the need for massive GPUs or infinitely scaling architectures.
The goal is to move past chatbots and tool-using agents toward something more persistent, auditable, and self-refining, all without relying on billions of parameters.
Top-level features of Redprint:
* Symbolic action-outcome chains (with execution feedback)
* Modular plug-ins for curiosity, compression, and evaluation
* Symbolic + vector memory fusion
* Tests that track agent learning over time
We’re not claiming AGI. But we are taking a real step toward agents that can reason about their actions and improve themselves.
Still early, but we wanted to surface the work now to find others exploring similar paths and avoid building in a vacuum.
Whitepaper coming soon. Select early access likely.
4 comments
[ 2.9 ms ] story [ 22.8 ms ] threadPlease chime in if you feel any interest in the subject. I'm not asking for money or hype or anything like that. Ideally someone who's ready to challenge what I have set forth and perhaps participate in the building.
We’ve been experimenting with a symbolic reasoning layer that sits beneath local LLMs, aimed at compressing their context demands by shifting some of the burden into structured memory and logic. The core idea: instead of infinitely scaling GPUs and tokens, we scaffold a learning loop that binds action to outcome, reflection to memory, and memory to improvement.
The architecture includes:
* A lightweight Prolog or Datalog engine to track symbolic execution paths
* Vector and symbolic memory fusion to keep both nuance and structure
* Outcome tests that update a learning score over time, enabling the agent to refine future actions
* Curiosity modules that bias the agent toward resolving ambiguity or closing feedback loops
One example: we had an agent running a simple multi-step task loop with an objective scoring function. First run: 0 percent success. But as outcome chains were logged and the reasoning engine updated its knowledge base, the same model climbed into the 70 percent range over 60 runs. No retraining, no fine-tuning, just structured feedback and symbolic state retention.
Still early, but the goal is not just to build a better chatbot or prompt wrapper. We’re aiming for something more like a persistent local intelligence that reasons through problems, remembers its missteps, and adapts without external retraining.