Research-grade cognitive AI architecture, built for the browser.
ReflexEngine V3 enables multi-layered AI reasoning with conscious/subconscious thought cycles, semantic memory recall, and self-narrative weaving—all running directly in your browser using Google's Gemini API.
A highly compact, graph-based memory system that models semantic transitions between concepts. The SRG:
- Extracts key concepts from each user and model interaction
- Learns weighted associations between cognitive states (conscious thoughts, subconscious reflections, axioms)
- Predicts follow-up concepts and enriches context for future queries
- Stores and retrieves patterns across sessions using IndexedDB
The SRG acts as the "synaptic network" of Reflex, enabling intelligent memory recall and semantic prediction without external vector databases.
Reflex uses a three-layer processing cycle:
- Conscious Thought: Initial, direct response to the user's query
- Subconscious Reflection: Deep introspection, critique, and alternative reasoning paths
- Final Synthesis: Integration of both layers into a polished, context-aware answer
Each cycle generates cognitive trace atoms that feed into the SRG and inform future reasoning.
After each interaction, an Arbiter synthesizes insights from the cognitive trace into axioms—distilled principles or lessons that capture the essence of the exchange. Over time, these axioms form a growing knowledge base specific to your conversation.
Reflex maintains an evolving Core Narrative, woven from insights and axioms generated during Arbiter cycles. This narrative acts as a coherent "self-story" that reflects the AI's accumulated understanding and priorities across all sessions.
- Import project files and selectively include them in the AI's context
- Toggle context for individual files, messages, and user/model turns
- Diff viewer for comparing files side-by-side
- Export generated files individually or as a zip archive
- Session state persistence: Save and load full conversation histories as JSON
A 3D force-directed graph visualization of the AI's internal memory:
- Nodes represent conscious thoughts, subconscious reflections, and axioms
- Links show semantic associations and concept co-activation
- Interactive timeline playback to watch memory formation in real time
- Zoom, pan, and hover to explore the structure of cognition
-
Prerequisites:
- Node.js (v18+ recommended)
- A Google AI API key
Steps:
- Clone the repository: