Vibe Coding Discoverdsh-memory
View on GitHub白箱AGI架构探索：元认知（自我认知循环）、持续学习（知识飞轮）、世界模型（条件空间+语义时空图）、自我改进（自举纪律）、零LLM白箱管线与可审计信任护栏。
Long-term memory for AI agents: conversations sink into plain-markdown cognitive graphs, with deterministic rule-based retrieval, guardrails and full audit trails (no LLM judgment, no vectors). Ships as a stdio MCP server / DSH-Cordis plugin for Claude Code, Codex CLI or CodeBuddy.
Use Cases
Give coding agents cross-session long-term memoryMount a stdio MCP memory server into Claude Code / Codex CLI / CodeBuddy / DSHAuto-capture conversation turns into an auditable markdown cognitive graphDeterministic rule-based recall and timeline reconstruction of past sessionsBenchmark memory retrieval with public LoCoMo-style datasetsGuardrail and audit agent writes (block, log, forget only on explicit request)Run concurrent multi-agent task scheduling with a Rust worker poolContext management: importance scoring, budget packing, layered injection
Built With
- Language
- Python
- Frameworks
- MCP · DeepSeek Harness (DSH) · Cordis · Claude Code · Codex CLI · CodeBuddy · Node.js · Rust · Python
Tags
long-term-memory · persistent-memory · knowledge-graph · mcp-server · multi-agent · rag · guardrails · agent-safety · explainable-ai · auditability · cognitive-graph · retrieval · self-improvement · world-model · deepseek · chinese-nlp