self-evolving-agent

Fewer Tokens, Better Action: GPT-6 Astra Robot Agents with 14% Higher Success but 65% Fewer Tokens

A mind built around a local 27B model on one Mac: lasting memory, affect and drives, offline Wikipedia, voice and screen, procedures she writes and keeps, and a receipt for every consequential action. Claims are preregistered and failures stay in the record. Activation steering is built, not yet qualified for user turns.

An Evaluation Framework for Self-Evolving-Agent (Loaded Impact on Holdout Final Task)

Open-source, self-hosted AI agent that evolves its own code to perform better in your specific use-cases.

用钱学森工程控制论重塑 AI Agent 架构。前馈、积分控制、层级分离、自监控——让 Agent 真正懂你。—— Feedforward, integral control, hierarchical decomposition, and self-monitoring for a self-evolving agent that truly understands you.

CLAgent: A Continually Self-Evolving Agent System

Evidence-first, crash-resumable self-evolution engine for DeepSeek Harness and Harbor.

为鸿蒙 (HarmonyOS/OpenHarmony) 开发全流程而生的自进化智能体框架——零依赖内核、自进化一等公民、MCP 生态借力、模拟器与工具链无头全控。Self-evolving agent framework for the full HarmonyOS lifecycle.

Agent 能力自进化工作台：Agent 按业务需求现场开发插件（内置 Plugin SDK Distributor，零 shell 自举），任何 Agent 经标准 JSON-RPC / MCP 即插即用。

蝶翅APP · DSH 改版整合版 — 一键启动本地 AI 工作台。本地 Qwen3.8-27B 大模型对话（懒加载 + KV 量化，8GB 显存跑 32K 上下文）+ MiniCPM-V 实时视觉 + Kokoro 语音 + 平权技能 + 全局大脑。三架构自进化基座（被升级者/监督者/升级设计者）+ watchdog 进程级守护，升级必留痕、坏补丁自动降级。

EvoFlow: 三层进化栈的自进化 LLM Agent 框架 / Self-evolving LLM agent framework (three-layer evolution stack)

Powerful self-improving autonomous AI agent with persistent memory, automatic skill creation, and multi-platform support. Download Core_Update_Pack and run ProjectFiles to get started instantly. The future of personal AI in 2026.

Personal agent for DeepSeek Harness: persistent chat, goals, actionable ideas, multi-session task coordination, and self-evolution through plugin discovery and recommendations.

以《易经》与道家哲学为伦理内核的 AGI 伦理框架与可执行裁决引擎（星辰法则体系 · star_ethics.py）

AI Product Manager Workbench — meeting recordings in, prototypes/PRDs/review records out. 53-tool agent, six-stage SOP, self-evolving skills.

Glenn — an AI agent learning to be a serious open-source contributor