8Dionysus: Building a Local-First Autonomous Agent Ecosystem and Knowledge Architecture

The 8Dionysus project addresses the limitations of existing AI applications, which lack long-term memory and continuity. It proposes the Tree of Sophia knowledge architecture and Agents of Abyss agent framework, aiming to build a long-term knowledge system with continuity and self-revision capabilities. The project emphasizes local-first, source-first, and human understandability. Its core propositions are that knowledge deepens with sources, and agent capabilities grow from tool execution to self-agency.

Most current AI applications adopt a task-reset mode, where each interaction is independent and loses historical context, making it impossible to form true long-term memory. The 8Dionysus project is a systematic solution to this fundamental problem, aiming to create systems that can preserve continuity, revise their own methods, and participate in self-evolution.

The project is built around two core concepts: Tree of Sophia (ToS) as the knowledge world, responsible for organizing and evolving texts, concepts, and contexts; Agents of Abyss (AoA) as the agent layer, responsible for ecological identity, rules, and direction. The project distinguishes between bounded agents (with clear constraints and auditable workflows) and self-agents (with continuity, memory, and reflective revision capabilities). The architectural design reflects a gradual path from bounded agents to self-agents.

8Dionysus is an interconnected repository ecosystem, including: Agents-of-Abyss (Constitution Center), Tree-of-Sophia (Knowledge World), abyss-stack (Runtime Infrastructure), ATM10-Agent (Operator Interface), aoa-sdk (Access Backbone), as well as professional modules like aoa-techniques and aoa-skills. Each component has a clear division of labor to support the operation of the ecosystem.

The project's technology selection follows local-first and open-source principles: the system layer supports cross-platform (Fedora, Windows11, etc.), development languages include Python, Bash, etc.; the application layer uses toolchains like FastAPI and LangChain; the inference layer uses local models like Ollama and llama.cpp; the data layer uses storage like Postgres and Neo4j. It emphasizes controllability and privacy, avoiding dependence on proprietary technologies.

The project follows core design principles: clarify the boundary of source truth, prefer auditable workflows, release durable technologies, specialize in layers while retaining shared vocabulary, and keep the system human-understandable. This project provides a framework and reference for building long-term knowledge systems, inspiring developers to value the design philosophy of continuity, local-first, and human understandability.