Valence Sphere

ValenceSphere, an experimental concept-first AI reasoning system, was released as open-source software under Apache 2.0 on 14 August 2026. The system uses a triadic structure of questioner, answerer, and adjudicator to audit and verify data, aiming to give AI models greater structure and critical thinking capabilities.

ValenceSphere is an experimental "concept-first" AI reasoning system, using triadic questioner Socrates , answerer Plato , and adjudicator for data auditing and verification to give AI models greater structure and the beginnings of critical thinking. It is structured as a two-stage concept-learning and factual-auditing system. Stage 1 builds structured knowledge about individual concepts. Stage 2 provides an ordinary LLM chat while independently checking factual assertions made by both the user and the LLM. The program does not use machine learning models of its own, just yet, but this can be integrated later. It coordinates configured LLM APIs, structured concept templates, persistent audit records and a navigable knowledge graph. Core purpose ValenceSphere is intended to: Program structure ValenceSphere has two clearly separated stages. Stage 1 — Concept Formation Stage 1 creates the knowledge that ValenceSphere later uses. Its tabs are: ValenceSphere/ └── lemon/ └── lemon.json The template is updated atomically. ValenceSphere does not create rollover copies or multiple versions of the same concept template. Stage 2 — Model Workspace Stage 2 opens in a separate window. The left side is an ordinary LLM chat. ValenceSphere does not inject its templates, audits or conclusions into the normal chat request. After each response appears, ValenceSphere independently scans: Summary ValenceSphere combines structured concept formation with transparent conversational fact-checking. Notes Valence Sphere is not fully perfected and finished yet, and will be subject to further revisions and edits to come, but was released on 14 August 2026 as open source software under Apache 2.0 license terms. The original concept was created on 30 July 2025. It has gone through several revisions since then. This program was built to create an alternate AI reasoning architecture that starts from the concept up. It uses LLMs as a scaffold to do that. The intended outcome is that the model will gain greater structure and become capable of its own internal reasoning and develop critical judgment of discrete facts. Through its auditing and verification process, it can determine how reliable its own knowledge is.