diff --git a/README.md b/README.md index 1e3ba2df..e7a4faf4 100644 --- a/README.md +++ b/README.md @@ -1,6 +1,6 @@
-![Raven banner](https://github.com/user-attachments/assets/6c6f585a-21b6-4e7b-9187-acffe59d0c10) +![Raven banner](https://github.com/user-attachments/assets/d56804e5-5d4b-4493-bc70-71bd38833806)

X @@ -17,13 +17,23 @@ # Raven -Raven is **The Self-Improving Agent Harness**, built on [EverOS](https://github.com/EverMind-AI/EverOS), with opt-in Deep Research for multi-source investigation. +Raven is the open-source, **self-improving Agent Harness** you can run today. It brings terminal-first execution, local tracing, long-term memory, skills, evaluation, and reusable workflows into one system for long-running AI work. -Raven helps agents improve across runs by continuously refining the systems around them: tools, skills, memory, code execution, policies, and working environment. EverOS provides durable user memory, agent memory, and world knowledge across sessions, so successful workflows can evolve into reusable Agent Templates and digital workers. +## The Harness of Harnesses -**Update:** Raven added Deep Research. Enable it with `raven deep-research enable` -to give the agent access to MiroThinker-backed, multi-source research when a -task needs deeper investigation. +As AI agents move from narrow tasks toward long-running, cross-domain work, manually designing a single, ever-larger harness stops scaling. A harness optimized for one model or domain also cannot provide every capability needed for general intelligence. + +**Raven** also names the broader research and ecosystem direction behind this work: **The Harness of Harnesses**. It aims to automatically build and improve Agent Harnesses for specific models and domains, then compose their heterogeneous execution capabilities into an **All-Domain Collaboration Network**. + +| **Trusted** | **Persistent** | **Evolving** | +| --- | --- | --- | +| Capability is earned through verified performance, not self-declared labels. | Verified results, task state, and long-term memory carry across executors. | Every real run improves capability profiles, skills, routing, and the network itself. | + +Raven does not treat a model and its harness as a fixed pair. Through a continuous **evaluation -> execution -> verification -> memory -> feedback** loop, it discovers, composes, and improves the right capabilities for each task. Validated work becomes reusable experience, allowing both individual agents and the wider capability network to evolve. + +The Raven evaluation spans **22 Agent benchmark tasks** across task quality, cost, and key mechanism gains. The reported results show broad performance and efficiency improvements over existing agent systems while advancing the **quality-cost Pareto frontier**. + +> Raven names both the runnable open-source harness in this repository and the broader multi-agent research and ecosystem direction described above. > Raven is pre-alpha. Interfaces and configuration may change quickly. diff --git a/README.zh-CN.md b/README.zh-CN.md index 37e3c5b5..638b4165 100644 --- a/README.zh-CN.md +++ b/README.zh-CN.md @@ -1,6 +1,6 @@

-![Raven banner](https://github.com/user-attachments/assets/6c6f585a-21b6-4e7b-9187-acffe59d0c10) +![Raven banner](https://github.com/user-attachments/assets/d56804e5-5d4b-4493-bc70-71bd38833806)

X @@ -17,16 +17,23 @@ # Raven -Raven 是构建在 [EverOS](https://github.com/EverMind-AI/EverOS) 之上的 -**The Self-Improving Agent Harness**，并内置可选 Deep Research，用于多来源深度研究。 +Raven 是今天即可运行的开源 **自我进化 Agent Harness**。它将终端执行、本地 Tracing、长期记忆、Skills、评测与可复用工作流整合进同一套系统，面向长程 AI 任务持续学习与改进。 -Raven 会持续迭代支撑 Agent 的 harness：tools、skills、memory、code execution -runtime、policies 和工作环境。EverOS 为这个 harness 提供跨会话持久存在的用户 -记忆、Agent 记忆和世界知识，让每一次运行都能改进 Agent 的行动方式、知识状态， -并把可重复工作流沉淀成可复用 Agent Templates 和 digital workers。 +## The Harness of Harnesses -**Update：** Raven 新增 Deep Research。运行 `raven deep-research enable` 后， -Agent 可以在需要深度调查的任务中使用 MiroThinker-backed、多来源 research tool。 +随着 AI Agent 从单一任务走向长程、多领域协作，依赖人工设计一个不断膨胀的 harness 已经难以持续扩展；而与特定模型和领域深度绑定的单一 harness，也无法覆盖通用智能所需的全部能力。 + +**Raven** 同时代表这一工作的更广阔研究与生态方向，其核心定义是 **The Harness of Harnesses**。它旨在自动构建和提升适配特定模型与领域的 Agent Harness，并将不同 Harness 的异构执行能力汇聚为统一的 **全领域协作网络（All-Domain Collaboration Network）**。 + +| **可信** | **可延续** | **可进化** | +| --- | --- | --- | +| 能力由真实、可验证的表现决定，而非自我声明。 | 经过验证的结果、任务状态与长期记忆能够跨执行者延续。 | 每次真实执行都会改进能力档案、Skills、调度与整个协作网络。 | + +Raven 不再把模型与 harness 视为静态组合，而是通过持续的 **评测 → 执行 → 验证 → 记忆 → 反馈** 闭环，发现、编排并优化每项任务所需的能力。经过验证的工作会沉淀为可复用经验，使 Agent 个体与更广泛的能力网络共同进化。 + +Raven 的评测覆盖 **22 个 Agent 基准任务**，从任务质量、成本与关键机制收益等维度进行验证。报告结果显示，其在性能与效率上相较现有 Agent 系统实现了全面提升，并进一步推进了 **质量—成本帕累托前沿**。 + +> Raven 既指本仓库中可运行的开源 harness，也指上述更广阔的 Multi-Agent 研究与生态方向。 > Raven 目前处于 pre-alpha 阶段，接口和配置可能快速变化。