The idea

OpenEvolve evolves code using fitness functions. But who decides the fitness criteria? Who verifies the evolved code is actually sound?

OpenBrainLM is a brain harness for LLMs that solves this with a Trinity dialectic — three voices (Ethos, Logos, Pathos, from Aristotle's Nicomachean Ethics) that debate every decision:

• Ethos — the evidence corridor. Holds the verified evidence base AND evaluation criteria.
• Logos — the evaluator. Only approves when criteria are validated against measurable parameters.
• Pathos — the mission. Argues for the goal, tries creative paths to get there.

Both sides must wield the same shared evidence to convince each other. No action proceeds until the fight resolves.

How it connects to OpenEvolve

From our concept paper:

1. You define the criteria (Ethos): what "better" means
2. The dialectic evaluates whether current code meets criteria (Logos checks, Pathos suggests improvements)
3. If the dialectic says this needs to be better → hand code + fitness function to OpenEvolve
4. OpenEvolve evolves the code — mutating, testing, selecting
5. Evolved code comes back through the dialectic for verification (did evolution break anything?)

The dialectic sets the goal. OpenEvolve finds the path. The dialectic verifies the result.

What else OpenBrainLM does

• Brainstem — hooks that fire before/after every action. Inhibition-by-default.
• Long-term memory — hippocampal consolidation (short-term → long-term), quarantine layer, schema-driven verification.
• Self-learning — the system evolves its own routing during sleep cycles using NEAT neuroevolution.
• 8-layer biomimetic architecture (roadmap) — every layer derived from real biology with citations.

LLM-agnostic, pure Python stdlib, MIT licensed.

Repo: https://github.com/0SxD/OpenBrainLM

The AlphaEvolve integration section in concept paper 1 was written with OpenEvolve in mind. Would love feedback from this community.