Self-Evolving · Human-on-the-Loop · End-to-End自进化 · 人在回路 · 端到端

EvoScientist

Your self-evolving AI scientist.你的自进化 AI 科学家。

Plan, research, code, analyze, and write with a multi-agent system that remembers your methods and improves across research runs.用一个多智能体系统完成规划、调研、编码、分析与写作——它记住你的方法，并在一次次研究运行中不断进化。

v0.2.8 Apache 2.0 Python 3.11+

Explore on GitHub在 GitHub 上探索 4.4k Quick Start快速开始 Technical Report技术报告

Research workspace研究工作区Connected已连接

Dynamic agent swarmSync · Async · On-demand动态智能体集群同步 · 异步 · 按需

EvoMemoryPersistent research knowledge graphEvoMemory持久化研究知识图谱

#1 DeepResearch Bench At submission time提交时

#1 AstaBench Code & data analysis代码与数据分析

Dynamic Agent orchestration动态智能体编排 Sync · async · on-demand同步 · 异步 · 按需

16 EvoSkills Ideation to publication从构思到发表

Latest最新

News & Updates新闻与更新

03 Jun 2026 🥈 Ranked #2 overall — and 🥇 #1 among GPT-5.4-based agents — on ResearchClawBench (Agent Mode)! Leaderboard 👈总榜排名第 2 —— 在基于 GPT-5.4 的智能体中夺得 🥇 第 1 —— 登上 ResearchClawBench（智能体模式）！排行榜 👈

18 Apr 2026 🥇 Ranked #1 on DeepResearch Bench at submission time! Leaderboard 👈在 DeepResearch Bench 上提交时排名第 1！排行榜 👈

13 Apr 2026 🥇 Reclaimed #1 on DeepResearch Bench II at submission time! Leaderboard 👈在 DeepResearch Bench II 上重夺提交时排名第 1！排行榜 👈

26 Mar 2026 🥇 Ranked #1 on AstaBench Data Analysis at submission time! Leaderboard 👈在 AstaBench Data Analysis 上提交时排名第 1！排行榜 👈

25 Mar 2026 🥇 Ranked #1 on AstaBench Code & Execution at submission time! Leaderboard 👈在 AstaBench Code & Execution 上提交时排名第 1！排行榜 👈

13 Mar 2026 🚀 EvoScientist officially debuts!EvoScientist 正式发布！

11 Mar 2026 ⛳ Technical Report is live! Check it out 👈技术报告已发布！点击查看 👈

06 Mar 2026 🥇 Ranked #1 on DeepResearch Bench II at submission time! Leaderboard 👈在 DeepResearch Bench II 上提交时排名第 1！排行榜 👈

24 Nov 2025 🏆 6/6 accepted at ICAIS 2025 AI Scientist Track — Best Paper & AI Reviewer's Appraisal Award! Details 👈在 ICAIS 2025 AI Scientist Track 上 6 篇论文全部被接收（6/6）—— 荣获最佳论文奖及 AI 审稿人评价奖！详情 👈

Releases发布记录

Changelog更新日志

v0.2.8 21 Aug 2026

Faster deploy-mode rebuilds & Novita AI. Deploy-mode graph rebuilds drop from ~15 s to under a second; resume commands no longer crash and empty session history; Novita AI as a new LLM provider; NVIDIA BioNeMo Agent Toolkit in onboarding's recommended skill packs; bounded routed reasoning with empty truncations surfaced as errors; a subscription OAuth recipe in the docs; deepagents 0.7.8.更快的 deploy 模式重建与 Novita AI。deploy 模式下的图重建耗时从约 15 秒降至 1 秒以内；resume 命令不再崩溃或清空会话历史；新增 Novita AI 作为 LLM 提供商；onboarding 推荐技能包中加入 NVIDIA BioNeMo Agent Toolkit；路由推理加入上限约束，空截断会以错误形式明确提示；文档新增订阅制 OAuth 使用指南；升级至 deepagents 0.7.8。

v0.2.7 14 Aug 2026

Faster startup & new models. Lazy imports, a reusable kept-alive langgraph dev server, and indexed thread listing; new models: Gemini 3.7 Flash (Google + OpenRouter), DeepSeek V4 Pro 0813, Grok 4.6, and GLM-5.3 (Zhipu + OpenRouter); encrypted-webhook channels reject unsigned POSTs; MCP stdio servers start reliably under redirected streams on Windows; sub-agent tools resolve at dispatch time; /compact history offload restored on deepagents 0.7.6.更快启动与新模型。启动提速：延迟导入、可复用的常驻 langgraph dev 服务器、索引化线程列表；新增模型：Gemini 3.7 Flash（Google 与 OpenRouter）、DeepSeek V4 Pro 0813、Grok 4.6 及 GLM-5.3（Zhipu 与 OpenRouter）；加密 webhook 渠道现在会拒绝未签名的 POST 请求；Windows 下 MCP stdio 服务器在重定向流环境中可稳定启动；子智能体工具改为在派发时解析；在 deepagents 0.7.6 上恢复了 /compact 历史卸载。

v0.2.6 07 Aug 2026

Agent teams. Invite installed expert skills into a session (/expert <name>) for in-turn consults, parallel panels, or background jobs; configurable bind hosts for the langgraph dev backend and WebUI (loopback by default); Volcengine Coding Plan provider (glm-5.2, kimi-k2.5); Qwen3.8-Max on DashScope and OpenRouter (1M context); deepagents 0.7.5 with media placeholders instead of provider 400s; a blank tool-call ID fix for Kimi/Zhipu sessions.智能体团队。可将已安装的专家技能邀请进会话（/expert <name>），用于回合内咨询、并行专家评审或后台任务；langgraph dev 后端与 WebUI 支持可配置的绑定主机（默认仅监听本机回环）；新增 Volcengine Coding Plan 提供商（glm-5.2、kimi-k2.5）；DashScope 与 OpenRouter 上新增 Qwen3.8-Max（100 万上下文）；升级至 deepagents 0.7.5，以媒体占位符替代提供商 400 报错；修复 Kimi/Zhipu 会话中工具调用 ID 为空的问题。

v0.2.5 01 Aug 2026

Unified HITL & new providers. Unified human-in-the-loop approval across the main agent and sync + async sub-agents; deepagents 0.7.0 with leaner built-in prompts and a recursive delete tool gated behind the same approval as execute; Requesty and Atlas Cloud as new LLM providers; fixes for unnamed tool calls, interrupted tool-call history on sync sub-agents, and deploy-mode port propagation.统一 HITL 审批与新提供商。主智能体与同步 + 异步子智能体现在共用统一的人在回路审批；升级至 deepagents 0.7.0，内置提示更精简，并新增递归 delete 工具（与 execute 共用同一审批门控）；新增 Requesty 与 Atlas Cloud 两个 LLM 提供商；修复了未命名工具调用、同步子智能体的工具调用历史中断，以及 deploy 模式端口传递问题。

v0.2.4 26 Jul 2026

Opus 5 & Gemini 3.6. Claude Opus 5 selectable on Anthropic and OpenRouter (incl. fast), plus Gemini 3.6 Flash and 3.5 Flash Lite on Google and OpenRouter; Kimi K3 now works over Anthropic-protocol channels (Kimi For Coding, custom endpoints), covering structured output, history replay, and multi-turn thinking; fixes for interrupted tool-call history, skill-install path leaks, and OpenRouter SSE streaming (pinned below 0.11).Opus 5 与 Gemini 3.6。Claude Opus 5 现可在 Anthropic 与 OpenRouter 上选用（含 fast 版本），另新增 Gemini 3.6 Flash 与 3.5 Flash Lite（Google 与 OpenRouter）；Kimi K3 现已支持 Anthropic 协议通道（Kimi For Coding、自定义端点），覆盖结构化输出、历史回放与多轮思考；修复了工具调用历史中断、技能安装路径泄漏，以及 OpenRouter SSE 流式传输问题（固定在 0.11 以下版本）。

v0.2.3 18 Jul 2026

Kimi K3 & provider fixes. Kimi K3 selectable on Moonshot and OpenRouter (1M context); async sub-agent runs no longer get stuck pending thanks to orphaned-run cleanup; Telegram slash commands; provider fixes (DeepSeek native SDK, GPT-5.x via ChatGPT OAuth, OpenAI reasoning_effort); quieter tool-selector streaming and smaller checkpoints.Kimi K3 与提供商修复。Kimi K3 现可在 Moonshot 与 OpenRouter 上选用（100 万上下文）；通过孤立运行清理机制，异步子智能体运行不再卡在 pending 状态；新增 Telegram 斜杠命令；多项提供商修复（DeepSeek 原生 SDK、通过 ChatGPT OAuth 使用 GPT-5.x、OpenAI reasoning_effort）；工具选择器流式输出更安静，检查点体积更小。

v0.2.2 11 Jul 2026

New models. Selectable in onboarding and /model: GPT-5.6 (sol, terra, luna) for OpenAI and OpenRouter, plus Grok 4.5 and Tencent Hunyuan HY3 on OpenRouter; tighter config-file permissions and a reworked onboarding OAuth flow for auxiliary models.新模型。onboarding 与 /model 中新增可选模型：OpenAI 与 OpenRouter 上的 GPT-5.6（sol、terra、luna），以及 OpenRouter 上的 Grok 4.5 与腾讯混元 HY3；收紧了配置文件权限，并重构了辅助模型的 onboarding OAuth 流程。

v0.2.1 05 Jul 2026

AutoSkills. EvoMemory drafts reusable skills from its own observation clusters for you to review via /autoskills; a new --output-format stream-json for headless / SDK clients; richer slash-command completions; Windows UTF-8 config reads; a TUI welcome-banner fix; langchain-openrouter 0.2.5.AutoSkills。EvoMemory 会从自身的观察聚类中起草可复用技能，供你通过 /autoskills 审阅；为无头模式 / SDK 客户端新增 --output-format stream-json；更丰富的斜杠命令补全；Windows UTF-8 配置读取；修复 TUI 欢迎横幅；升级 langchain-openrouter 至 0.2.5。

v0.2.0 26 Jun 2026

Autonomy milestone. Scheduled tasks: cron-style recurring runs you set up with /schedule or natural language, running unattended with shell-access gating for safety; self-linking memory that connects related observations into a knowledge graph (complements / contradicts / supersedes) as it grows; a read-only GET /api/models endpoint exposing the model registry to the WebUI picker; the main agent now always keeps its memory tools available; deps: deepagents 0.6.12 / langchain-quickjs 0.3.2.自主性里程碑。计划任务：通过 /schedule 或自然语言设置类 cron 的周期性运行，无人值守运行并通过 shell 访问权限门控确保安全；自关联记忆会随着增长将相关观察连接成知识图谱（互补 / 矛盾 / 取代）；新增只读端点 GET /api/models，向 WebUI 模型选择器暴露模型注册表；主智能体现在始终保持记忆工具可用；依赖升级：deepagents 0.6.12 / langchain-quickjs 0.3.2。

v0.1.9 23 Jun 2026

Hotfix for fresh installs: the first message crashed with The subagent task tool cannot be exposed via ptc after deepagents 0.6.11 / langchain-quickjs 0.3 reserved task as the REPL global. Removed task from the code-interpreter PTC allowlist (task() stays available as the REPL global; async dispatch stays in PTC) and pinned deepagents[quickjs]~=0.6.11.针对全新安装的热修复：在 deepagents 0.6.11 / langchain-quickjs 0.3 将 task 保留为 REPL 全局变量后，首条消息会因 The subagent task tool cannot be exposed via ptc 而崩溃。已将 task 从代码解释器 PTC 白名单中移除（task() 仍作为 REPL 全局变量可用；异步分发仍保留在 PTC 中），并固定 deepagents[quickjs]~=0.6.11。

v0.1.8 22 Jun 2026

LangGraph gateway layer: UI-agnostic graph & thread access shared across CLI / TUI / serve / channel; OpenRouter Anthropic prompt caching now on by default (opt out with openrouter_anthropic_prompt_cache=false); slash-command Enter now submits correctly when a command name prefixes another; pre-commit ruff bump.LangGraph 网关层：与 UI 无关的图与线程访问，在 CLI / TUI / serve / channel 之间共享；OpenRouter 的 Anthropic 提示缓存现在默认开启（可通过 openrouter_anthropic_prompt_cache=false 关闭）；当命令名是另一命令的前缀时，斜杠命令 Enter 键现在能正确提交；升级 pre-commit ruff。

v0.1.7 16 Jun 2026

Memory retrieval: agents run a per-task preflight over stored observations (search_observations ranked keyword search + read_memory); multi-stage slash-command completions with subcommand awareness; Windows reliability fixes (async MCP tool execution + graph-state recovery after interruptions, cmd.exe path quoting); quoted virtual-path handling; deepagents 0.6.10.记忆检索：智能体在每个任务开始前对已存储的观察进行预检（search_observations 关键词排序搜索 + read_memory）；支持子命令感知的多阶段斜杠命令补全；Windows 可靠性修复（异步 MCP 工具执行 + 中断后的图状态恢复、cmd.exe 路径引用）；带引号的虚拟路径处理；升级 deepagents 至 0.6.10。

v0.1.6 11 Jun 2026

Session persistence fix: WebUI / langgraph dev threads survive restarts (SQLite checkpointer + scoped thread restore), memory-worker checkpoint cleanup (delete-on-completion + startup purge), short thread IDs in /threads and resume hints.会话持久化修复：WebUI / langgraph dev 线程可在重启后存活（SQLite 检查点 + 作用域线程恢复）、记忆工作进程检查点清理（完成后删除 + 启动时清理）、/threads 中的短线程 ID 与恢复提示。

v0.1.5 11 Jun 2026

Dangerous mode (real-filesystem access with safety checks), LangGraph streaming v3 pipeline, opt-in Anthropic prompt caching via OpenRouter, claude-fable-5, free-scrolling TUI, Windows CI support, public Cloudflare tunnel for EvoSci deploy (--tunnel).危险模式（带安全检查的真实文件系统访问）、LangGraph 流式传输 v3 管线、通过 OpenRouter 可选启用的 Anthropic 提示缓存、claude-fable-5、可自由滚动的 TUI、Windows CI 支持、为 EvoSci deploy 提供的公共 Cloudflare 隧道（--tunnel）。

v0.1.4 07 Jun 2026

Auxiliary model for background tasks & tool selection, observation-memory lifecycle, Qwen3.7-Max/Plus (DashScope), UI-backend selection, plus an OpenRouter multi-turn reasoning fix.用于后台任务与工具选择的辅助模型、观察记忆生命周期、Qwen3.7-Max/Plus（DashScope）、UI 后端选择，以及 OpenRouter 多轮推理修复。

v0.1.3 03 Jun 2026

Multimodal handling (image + PDF/doc flatten/hoisting, text-only model fallback), runtime-context middleware, memory middleware → profile files with stream timeline narration, textual CJK-input fix.多模态处理（图像 + PDF/文档扁平化 / 提升处理、纯文本模型回退）、运行时上下文中间件、记忆中间件 → 带流式时间线叙述的画像文件、textual 中日韩文输入修复。

v0.1.2 02 Jun 2026

Browser WebUI mode (beta), EvoSci deploy standalone LangGraph server, default model → claude-sonnet-4-6, MiniMax M3, plus sandbox-timeout and async-notifier channel-routing fixes.浏览器 WebUI 模式（beta）、EvoSci deploy 独立 LangGraph 服务器、默认模型 → claude-sonnet-4-6、MiniMax M3，以及沙箱超时与异步通知渠道路由修复。

v0.1.1 19 May 2026

deepagents 0.6.2 DeltaChannel upgrade, tier-aware skill mounts, status & elapsed-time bar, QQ inline buttons.deepagents 0.6.2 DeltaChannel 升级、分层感知技能挂载、状态与耗时进度条、QQ 内联按钮。

v0.1.0 08 May 2026

Async sub-agents (langgraph dev), official Docker image, personal WeChat, sessions-DB compaction.异步子智能体（langgraph dev）、官方 Docker 镜像、个人微信、会话数据库压缩。

v0.0.9 26 Apr 2026

Faster startup, in-session model switching, unified slash commands, DeepSeek V4 thinking fix.更快的启动速度、会话内模型切换、统一的斜杠命令、DeepSeek V4 思考修复。

v0.0.8 21 Apr 2026

Unified data directory, status bar, enhanced ask-user & auto-mode.统一数据目录、状态栏、增强的 ask-user 与自动模式。

v0.0.7 10 Apr 2026

Global skills directory, Moonshot/Kimi providers, ccproxy fixes, channel improvements.全局技能目录、Moonshot/Kimi 提供商、ccproxy 修复、渠道改进。

v0.0.6 03 Apr 2026

Dynamic context management, OpenRouter reasoning, More Effort mode, GLM-5.1.动态上下文管理、OpenRouter 推理、More Effort 模式、GLM-5.1。

v0.0.5 27 Mar 2026

Context-retry middleware, OpenAI relay config, Feishu event-loop fix, /compact.上下文重试中间件、OpenAI 中继配置、飞书事件循环修复、/compact。

v0.0.4 24 Mar 2026

@file mentions, resume history, Feishu WebSocket, LaTeX setup.@file 提及、恢复历史记录、飞书 WebSocket、LaTeX 环境配置。

v0.0.3 20 Mar 2026

Voice input (STT), MiniMax/DeepSeek providers, MCP & skill browsers.语音输入（STT）、MiniMax/DeepSeek 提供商、MCP 与技能浏览器。

v0.0.2 17 Mar 2026

OAuth sign-in, human-in-the-loop & ask_user, headless serve mode.OAuth 登录、人在回路与 ask_user、无头 serve 模式。

v0.0.1 13 Mar 2026

First public release of the self-evolving AI Scientist.自进化 AI 科学家首次公开发布。

Recognition荣誉

Awards & Benchmarks奖项与基准测试

Best Paper Award最佳论文奖

ICAIS 2025 · AI Scientist TrackICAIS 2025 · AI 科学家赛道

6/6 papers accepted. Best Paper & AI Reviewer's Appraisal Award. Details →6/6 篇论文被接收。最佳论文奖与 AI 审稿人评价奖。详情 →

AI-Generated Best PaperAI 生成最佳论文

ICAIS 2025

AI-authored research with end-to-end scientific quality. Details →由 AI 撰写、具备端到端科研质量的研究。详情 →

#1 DeepResearch Bench II

AgentResearchLab · Apr 2026

Reclaimed #1 on DeepResearch Bench II at submission time. Leaderboard →提交时重夺 DeepResearch Bench II 第 1。排行榜 →

#1 DeepResearch Bench

DeepResearch Bench · Apr 2026

Ranked #1 on DeepResearch Bench at submission time. Leaderboard →提交时在 DeepResearch Bench 排名第 1。排行榜 →

#1 AstaBench Code & Execution

AllenAI AstaBench · Mar 2026

Ranked #1 on AstaBench Code & Execution at submission time. Leaderboard →提交时在 AstaBench Code & Execution 排名第 1。排行榜 →

#1 AstaBench Data Analysis

AllenAI AstaBench · Mar 2026

Ranked #1 on AstaBench Data Analysis at submission time. Leaderboard →提交时在 AstaBench Data Analysis 排名第 1。排行榜 →

Quick Start快速开始

Up and running
in just seconds几秒之内
即刻上手

One wizard configures everything — LLM provider, API keys, model selection, and workspace mode. Supports OAuth sign-in for Claude Code and Codex CLI users.一个向导即可配置好一切——LLM 提供商、API 密钥、模型选择与工作区模式。支持 Claude Code 与 Codex CLI 用户的 OAuth 登录。

Install via uv tool install EvoScientist通过 uv tool install EvoScientist 安装
Run EvoSci onboard and follow the prompts运行 EvoSci onboard 并按提示操作
Choose your LLM provider and enter your API key选择你的 LLM 提供商并输入 API 密钥
Pick a model and workspace mode — done选好模型与工作区模式——完成

~ — EvoSci onboard

See It In Action实机演示

Demo演示

The same EvoScientist agent — driven from a desktop WebUI, the terminal, or your phone — plus its self-evolving EvoMemory graph.同一个 EvoScientist 智能体——通过桌面 WebUI、终端或手机驱动——外加它自进化的 EvoMemory 图谱。

Desktop WebUI桌面 WebUI

CLI / TUI

Mobile手机

EvoMemory · Knowledge GraphEvoMemory · 知识图谱

System Design系统设计

The self-evolving research loop自进化的研究循环

Every research run makes the system better: agents produce new findings, memory learns automatically, and successful methods evolve into reusable skills for the next run.每一次研究运行都让系统变得更好：智能体产出新发现，记忆自动学习，成功的方法进化为可复用的技能，服务于下一次运行。

flowchart LR
  GOAL["Research Goal"] --> AGENTS["Dynamic Multi-Agent Research"]
  AGENTS --> RESULTS["New Knowledge · Experiments · Reports"]
  RESULTS -->|"learns from every run"| MEMORY["EvoMemory<br/>Updates automatically"]
  MEMORY --> SKILLS["Evolving Skills<br/>Strengthen the next run"]
  SKILLS -->|"better ways to research"| AGENTS
  classDef evolve fill:#101a3e,stroke:#00e5ff,stroke-width:2px,color:#e8edf5
  class AGENTS,RESULTS,MEMORY,SKILLS evolve

Dynamic collaboration动态协作No fixed agent count — the right team forms around each research goal.没有固定的智能体数量——合适的团队围绕每个研究目标动态组建。

Memory that keeps learning持续学习的记忆Findings, preferences, and context update automatically across research runs.发现、偏好与上下文在一次次研究运行中自动更新。

Skills that self-evolve自进化的技能Successful methods become reusable skills that improve future work.成功的方法沉淀为可复用的技能，提升未来的工作。

Dynamic Agent System动态智能体系统

Orchestrated for every phase
of the scientific process为科研的每个阶段
精心编排

From hypothesis to publication — the main agent composes the right workers, tools, and skills for each research task.从假设到发表——主智能体为每个研究任务组合合适的工作者、工具与技能。

Dynamic Multi-Agent System动态多智能体系统

The main agent composes synchronous and asynchronous workers on demand from specialized agents, general-purpose agents, Skills, and MCP tools — with no fixed agent count.主智能体按需从专用智能体、通用智能体、技能与 MCP 工具中组合同步与异步的工作者——没有固定的智能体数量。

Persistent Memory持久化记忆

Context, preferences, and findings survive across sessions — self-linking into a knowledge graph (complements / contradicts / supersedes) so the system internalizes scholarly taste and builds on prior work.上下文、偏好与发现跨会话留存——自动关联成知识图谱（补充 / 矛盾 / 取代），让系统内化学术品味并在既有工作上持续积累。

Literature Research文献调研

Deep web search with structured 7-dimension reflection. Finds papers, methods, and baselines with enforced citation rigor.结合结构化 7 维反思的深度网络检索。以严格的引用规范查找论文、方法与基线。

Code Generation & Debug代码生成与调试

Write, execute, and iteratively debug experiment code in a sandboxed workspace with 300s timeout, output limits, and auto-recovery.在沙箱工作区中编写、执行并迭代调试实验代码，具备 300 秒超时、输出限制与自动恢复。

Scientific Workflow科研工作流

6-phase process: Intake → Plan → Execute → Evaluate → Write → Verify. Baseline-first design with one-variable iteration for scientific rigor.六阶段流程：接收 → 规划 → 执行 → 评估 → 撰写 → 验证。基线优先设计，单变量迭代，确保科学严谨。

MCP & EvoSkillsMCP 与 EvoSkills

Plug in MCP servers or install research-lifecycle skills from GitHub. Compatible with other AI coding agents out of the box.接入 MCP 服务器，或从 GitHub 安装覆盖研究全生命周期的技能。开箱兼容其他 AI 编码智能体。

Async Sub-Agents异步子智能体

Long-running tasks dispatched to background workers via start_async_task. The main agent stays responsive and is notified on completion — true parallel multi-agent execution.长耗时任务通过 start_async_task 派发给后台工作者。主智能体保持响应，并在完成时收到通知——真正的并行多智能体执行。

Adaptive Context自适应上下文

Per-turn tool filtering keeps only relevant tools visible. Dynamic system-prompt rewriting trims to conversation state. Automatic context overflow recovery prevents runaway sessions.逐轮工具过滤只保留相关工具可见。动态改写系统提示词，贴合对话状态精简。自动上下文溢出恢复，防止会话失控。

Human Control人工控制

Approve high-risk tool calls before they run. Voice input via faster-whisper (zh/en/auto). Rich slash commands — /threads, /resume, /schedule, and more.在高风险工具调用运行前进行审批。通过 faster-whisper（zh/en/auto）语音输入。丰富的斜杠命令——/threads、/resume、/schedule 等。

Scheduled & Autonomous定时与自主

Cron-style recurring runs via /schedule or natural language, executing unattended with shell-access gating. AutoSkills lets EvoMemory draft reusable skills from its own observation clusters.通过 /schedule 或自然语言进行 Cron 式定时循环运行，在 shell 访问门控下无人值守执行。AutoSkills 让 EvoMemory 从自身的观察聚类中起草可复用技能。

Deploy & Integrate部署与集成

Ship a standalone LangGraph server with EvoSci deploy — full MCP and async sub-agents over HTTP for external UIs and SDK clients, optionally exposed through a public Cloudflare tunnel.用 EvoSci deploy 部署独立的 LangGraph 服务器——通过 HTTP 向外部 UI 与 SDK 客户端提供完整的 MCP 与异步子智能体，可选经由公共 Cloudflare 隧道暴露。

Multimodal Intake多模态输入

Drop in images, PDFs, and documents — flattened and hoisted for analysis, with automatic text-only fallback when a model lacks vision.拖入图片、PDF 与文档——自动展平并提取以供分析；当模型不具备视觉能力时自动回退为纯文本。

Multi-Provider多提供商

Your models, your choice你的模型，你做主

Multiple LLM providers supported. One config to switch. Auto-detect model names or specify full IDs directly.支持多家 LLM 提供商。一处配置即可切换。自动识别模型名，或直接指定完整 ID。

A Anthropic

claude-opus-4-6claude-sonnet-4-6claude-haiku-4-5

O OpenAI

gpt-4oo3-minio1

G Google

gemini-2.5-progemini-2.5-flashgemini-2.0-flash

N NVIDIA

deepseek-r1llama-3.3-70bnemotron-ultra

S SiliconFlow

deepseek-v3qwen-plusglm-4

R OpenRouter

any model任意模型unified API统一 APIauto-routing自动路由

V Volcengine

doubao-prodoubao-lite

D DashScope

qwen-maxqwen-turbo

⚙ Ollama & CustomOllama 与自建

local / self-hosted本地 / 自建

ollama: prefixcustom base_url

Multi-Channel多渠道

One agent, every surface一个智能体，触达每个界面

CLI as the hub. Multiple messaging integrations — one agent session, any device.以 CLI 为中枢。多个消息渠道集成——同一个智能体会话，任意设备。

💬

iMessage

✈️

🎮

Discord

💼

Slack

💚

WeChat微信

🔔

DingTalk钉钉

🪶

Feishu飞书

📧

Email邮箱

🐧

🔒

Signal

Official Skill Repository官方技能仓库

EvoSkills

16 research-lifecycle skills covering the full pipeline from ideation to publication. Install all with a single command. Also compatible with Claude Code, Cursor, and other AI coding agents.16 个覆盖研究全生命周期的技能，贯穿从构思到发表的完整流程。一条命令即可全部安装。同样兼容 Claude Code、Cursor 及其他 AI 编码智能体。

research-ideation research-survey paper-navigator experiment-pipeline experiment-craft experiment-iterative-coder paper-planning paper-writing paper-review paper-rebuttal paper-figures academic-slides nano-banana paper-graph evomath-tao evo-memory

EvoScientist/EvoSkills

Skills技能

Stay tuned.
Big things are coming.敬请期待。
大事将至。

Benchmarks · More agents · EvoSkills v2基准测试 · 更多智能体 · EvoSkills v2

EvoScientist.ai@gmail.com Join Discord加入 Discord

Founder创始人

Xi Zhang张曦

Founder of EvoScientistEvoScientist 创始人

EvoScientist is a living research system built on a single conviction: full research autonomy is coming — but it won't come from a pre-built product. It will emerge from your own harness — a system that absorbs your judgment, encodes your taste, and compounds with every run. I call this Harness Vibe Research, and formalized its foundations as the Three Laws of Self-Evolving AI Agents — a framework for adaptive, collaborative, and lifelong agentic systems introduced in A Comprehensive Survey of Self-Evolving AI Agents: A New Paradigm Bridging Foundation Models and Lifelong Agentic Systems. The question was never whether this shift is coming — it's whether you're building the harness to grow with it.EvoScientist 是一个持续生长的研究系统，建立在一个信念之上：完全的研究自主正在到来——但它不会来自一个预制的产品，而会从你自己的 harness 中生长出来——一个吸收你的判断、编码你的品味、并随每一次运行不断复利的系统。我称之为 Harness Vibe Research，并将其基础形式化为 自进化 AI 智能体的三大定律——一个面向自适应、协作与终身智能体系统的框架，提出于 A Comprehensive Survey of Self-Evolving AI Agents: A New Paradigm Bridging Foundation Models and Lifelong Agentic Systems。问题从来不是这场转变是否会到来——而是你是否正在构建那个与之共同成长的 harness。

Homepage主页