Output specification for skill execution. Attributes: return_value: Return value from function execution. stdout: Standard o

Base data source error class

An agent designed to run LLM-based tasks with support for tools, memory, planning, callbacks, and automatic skill execution. This class

Base Class for OpenAI SDK LLMs. This class provides the base implementation for interacting with OpenAI-compatible models, including support

Detailed information about an import statement

Data not found exception

Result of executing a single skill. Attributes: skill_id: Identifier of the executed skill. success: Whether execution was s

Input specification for skill execution. Attributes: args: Command line arguments or positional parameters. kwargs: Keyword

Result of AutoSkills run containing the skill execution DAG. Attributes: dag: Adjacency list representation of skill dependencies.

A class representing a search request to ArXiv.

Search engine using arXiv API. Best for: scientific literature, research papers, preprints. Supports sorting by relevance, submission da

Automatic skill retrieval and DAG construction for user queries. Uses hybrid retrieval (dense + sparse) to find relevant skills, with LL

MCP client for all mcp tools This class can hold multiple mcp servers. Args: config(`DictConfig`): The config instance. mcp_

Executor for skill DAG with dependency-aware parallel execution. Handles execution order parsing, input/output linking between skills, a

Secure container for executing skills. Supports two execution modes: - use_sandbox=True: Execute in Docker sandbox via ms-enclave (recom

Unified web search tool for agents. It can search the web and fetch page content. - Search via multiple engines (Exa, SerpAPI, Arxiv) - D

ArXiv search result implementation.

Model for tracking research progress.

Base class for search responses.

Result of content summarization.

A ThreadPoolExecutor whose worker threads are daemon threads. Why: - In this repo, we run synchronous network calls in `run_in_executor`

Evidence management tool for DeepResearch agents. Provides structured storage and retrieval of evidence cards (notes). Each note represe

A file system operation tool

Response model for processed search results.

Model for research results.

A class representing the base fields of a search result. Attributes: url (str): The URL of the search result. id (str): The tempo

Tool for executing Python code in an isolated Docker sandbox. Features: - Complete Docker container isolation - Resource limits (mem

Integrated content optimization combining summarization and reranking. This is the main entry point for content optimization in the web sear

Configuration for content optimization.

Result of executing the entire skill DAG. Attributes: success: Whether all skills executed successfully. results: Dict mappi

Manages comments in unpacked Word documents.

Search engine using Exa API. Best for: semantic understanding, finding similar content, recent web pages with date filtering.

Fetch content using Jina Reader.

Represents key information extracted from a document. text (str): The text of the key information. Note that this text may contain `

LlamaIndexRAG class to implement the RAG of llama-index The configuration needed in the config yaml: - name: LlamaIndexRAG - embe

Mock class for arxiv.Result to simulate search results.

A workflow for conducting deep research tasks using LLMs and various tools.

Overview ------- A deep-research workflow that orchestrates recursive web search, structured extraction, and final answer/report gene

Search engine using SerpApi service. Best for: general web search via Google/Bing/Baidu, current events, news, and real-time information

A class representing a search request to SerpApi.

Represents a file within a Skill directory. Attributes: name: File name (e.g., "SKILL.md", "script.py") type: File extension

Complete schema for a Skill directory. Attributes: skill_id: Unique identifier for the Skill name: Skill name (max 64 charac

Specification for an AI agent's task planning and execution.

A tool special for task splitting

Todo list tool for ms-agent. Storage format: - {output_dir}/plan.json: { "schema_version": 1, "updated_at":

Interacting with Agent class, hold all tools

AKShare Data Source Implementation Supports multi-market data: - A-share market data - HK stock market data - US stock market da

Dynamically adjusts rate limits based on request success rate and error types. Automatically reduces request rate when rate limit errors are

Runs ms-agent as a subprocess with output streaming

Expose existing ms-agent agents as callable tools.

The app webui class.

Sparse retriever based on BM25 algorithm.

BaoStock Data Source Implementation Provides A-share market data: - Historical K-line data - Stock basic information - Financial

Thread-safe BaoStock session manager with connection reuse

Manage cooperative cancellation flags for FinResearch tasks (per user).

A workflow implementation that executes tasks in a sequential chain.

A code class can be executed in a `CodeAgent` in a workflow

Manages global configuration for the Web UI

Manages WebSocket connections

Summarize webpage content using a fast LLM model. This class provides asynchronous content summarization with: - Timeout protection -

Validator for Word document XML files against XSD schemas.

Workflow supporting multiple `next` tasks (DAG) with dynamic task execution.

XMLEditor that automatically applies RSID, author, and date to new elements. Automatically adds attributes to elements that support them when ins

A sandbox environment for securely executing code and commands based on `ms-enclave`. See `https://github.com/modelscope/ms-enclave`

A single execution record in the spec log. Attributes: execution_id: Unique identifier for this execution. skill_id: The ski

Specification log for tracking execution flow across skills. Attributes: spec_id: Unique identifier for this spec. title: Ti