核心功能

InnoClaw 提供了丰富的 AI 驱动研究和文件管理功能。

工作空间管理

工作空间将服务器端文件夹映射到应用程序中。每个工作空间是文件、对话历史和笔记的持久化容器。

  • 从任何允许的根目录（通过 WORKSPACE_ROOTS 配置）创建工作空间

  • 从首页打开已有的工作空间

  • 删除工作空间（仅删除数据库记录；磁盘上的文件会被保留）

  • GitHub 克隆 —— 直接将 GitHub 仓库克隆为新的工作空间

文件浏览器

文件浏览器提供工作空间内的完整文件管理功能：

  • 树形视图 —— 支持展开/折叠的层级目录浏览

  • 上传 —— 拖放或点击上传文件

  • 创建 —— 新建文件和文件夹

  • 编辑 —— 在内置编辑器中打开文件

  • 重命名 / 删除 —— 右键上下文菜单

  • Preview — Preview images, PDFs, 3D models (STL, OBJ, GLTF, etc.), and molecular structures (MOL)

  • 同步 —— 触发 RAG 索引以启用 AI 搜索

RAG 对话

AI 对话界面使用检索增强生成技术，基于您的工作空间文件回答问题：

  1. 同步工作空间以将文件索引到向量存储中

  2. 在对话面板中提问

  3. 接收回答，附带 [Source: filename] 引用指向相关文件

  4. 多轮对话 —— 上下文在消息之间保持连续

RAG 支持的文件类型

以下文件类型支持文本提取和索引：

分类

扩展名

文档

.pdf、.txt、.md、.html

数据

.json、.csv

代码

.js、.ts、.py、.java、.go、.rs、.cpp、.c、.rb、.php 等

笔记生成

从工作空间内容自动生成结构化笔记：

类型

描述

摘要

关键要点的简明概述

常见问题

包含答案的常见问题

简报

高管风格的简报文档

时间线

按时间顺序排列的事件

Memory

Agent conversation memory summaries

Daily Report

Auto-generated daily workspace activity report

Weekly Report

Auto-generated weekly workspace activity report

笔记也可以手动创建和编辑。

Agent 模式

Agent 模式提供一个可访问系统工具的自主 AI 助手：

Tool

描述

bash

Execute shell commands (configurable timeout up to 300s)

readFile

Read file contents within workspace

writeFile

Create or overwrite files

listDirectory

List directory contents with metadata

grep

Search file contents with regular expressions

searchArticles

Search arXiv and Hugging Face Daily Papers

getSkillInstructions

Load SCP scientific skill workflows

kubectl

Kubernetes cluster management (read-only by default)

submitK8sJob

Submit Volcano K8s jobs to Ascend 910B NPU clusters

collectJobResults

Collect K8s job logs and status

Agent mode supports three sub-modes:

Mode

Tools Available

Use Case

Agent

All tools

Full autonomous execution

Plan

readFile, listDirectory, grep

Read-only analysis and planning

Ask

readFile, listDirectory, grep

Simple question answering

技能

Skills are reusable AI workflow templates:

  • 206 built-in SCP (Science Context Protocol) skills from the Intern-Discovery Platform, covering 8 research domains:

    • Drug Discovery & Pharmacology (71 skills)

    • Genomics & Genetic Analysis (41 skills)

    • Protein Science & Engineering (38 skills)

    • Chemistry & Molecular Science (24 skills)

    • Physics & Engineering Computation (18 skills)

    • Experimental Automation & Literature Mining (7 skills)

    • Earth & Environmental Science (5 skills)

    • Cross-domain utility skills (2 skills)

  • Create custom skills with specific system prompts and instructions

  • Configure allowed tools per skill to control access

  • Define parameters with {{paramName}} template injection

  • Define multi-step workflows with tool hints per step

  • Import/export skills as JSON for sharing

Skills can be triggered from the Agent panel or auto-matched based on user intent.

Article / Paper Search

Search and study academic papers directly from the application:

  • Search Sources — arXiv and Hugging Face Daily Papers

  • Filter by keywords, date range, and source

  • Related Articles — Find papers related to a given article

  • Paper Study UI — Dedicated interface for searching, fetching, summarizing, and chatting about papers

  • Agent Integration — The searchArticles tool is available in Agent mode

Scheduled Tasks

Automate recurring operations with cron-based scheduled tasks:

Task Type

描述

daily_report

Generate daily workspace activity report

weekly_report

Generate weekly workspace activity report

git_sync

Automatically pull Git repositories

source_sync

Re-index workspace files

custom

User-defined workflows

Tasks can be global or workspace-specific, with execution tracking (status, errors, last run time).

Dataset Management

Download and manage datasets from HuggingFace Hub and ModelScope:

  • Download datasets, models, or spaces from HuggingFace or ModelScope

  • Import existing local directories as datasets

  • Lifecycle — Pause, resume, cancel, and retry downloads

  • Progress Tracking — Real-time download progress monitoring (0–100%)

  • File Preview — Preview dataset contents with split selection

  • Workspace Linking — Link datasets to workspaces (many-to-many)

  • File Filtering — Specify allow/ignore patterns for selective downloads

Cluster / Kubernetes Integration

Monitor and manage Kubernetes clusters for GPU-accelerated workloads:

  • Cluster Status — View nodes, jobs, and pods at a glance

  • kubectl / vcctl — Execute cluster commands via Agent mode

  • Volcano Job Submission — Submit jobs to Ascend 910B NPU clusters

  • Result Collection — Collect job logs, status, and exit codes

  • Audit Trail — Full operation history with workspace-level filtering

Requires KUBECONFIG_PATH and related K8S_* environment variables to be configured.

多 LLM 支持

在不同的 AI 提供商和模型之间切换：

提供商

模型

OpenAI

GPT-5.2, GPT-4.1, GPT-4.1 Mini, GPT-4.1 Nano, GPT-4o, GPT-4o Mini, o3, o3 Mini, o4 Mini

Anthropic

Claude Opus 4.6, Claude Sonnet 4, Claude 3.7 Sonnet, Claude 3.5 Haiku

Gemini

Gemini 2.5 Flash, Gemini 2.5 Pro, Gemini 3 Flash, Gemini 3 Pro, Gemini 3.1 Pro (Thinking)

Custom endpoints are supported via OPENAI_BASE_URL, ANTHROPIC_BASE_URL, and GEMINI_BASE_URL for third-party providers or self-hosted models.

双语界面

用户界面同时支持中文和英文：

  • 点击顶部导航栏中的语言切换按钮

  • 所有 UI 元素、标签和消息均已翻译

  • 语言偏好在会话之间持久保存

暗色模式

在浅色和深色主题之间切换：

  • 点击顶部导航栏中的主题切换按钮

  • 应用于所有面板，包括文件浏览器、对话和编辑器

  • 主题偏好在会话之间持久保存