Documentation

Start here, then dive into the features you need.

Getting Started

Quick Start

Deploy ScienceDiscovery

Concepts: how ScienceDiscovery completes a research task

Core Capabilities

ScienceDiscovery Core Capabilities

RSI for scientific artifacts: put each improvement to the test

Scientific execution environment and workspaces: turn ideas into runnable experiments

Idea Tree: explore research directions, then refine them

Scientific MCP and Skills: connect resources and reuse research methods

ScienceMemory and Reviewer: make conclusions easier to question

Specialists: give research tasks focused expertise

Domain Guides

Use PUCT to optimize a text compression algorithm

Research how migrating birds determine location and direction

Analyze correlations and clusters of sepsis endotype scores

Design an antibody on Ascend NPU

Advanced Setup

Configure custom MCP servers, OAuth and Inspector

Configure the Network Proxy

Install Neo4j and configure ScienceMemory

Import and manage research Skills

Create and use custom Specialists

Reference

Configuration, Ports, Quotas, and Storage Reference

REST API Reference

Runtime Behavior Reference

Built-in Tools Visible to the Model

Web Search and Web Fetch

JiuwenSwarm migration: status and hand-over

Agent backends

Built-in Research Capabilities Reference

CLI Reference

Shell, environments and workspaces

Developer Documentation

Developer Documentation

Runtime Architecture

Control Plane: services/api

Sandbox Execution: services/runner

ScienceMemory (Task and Citation Chains)

Subagent Orchestration and Governance

Skill Progressive Disclosure

Scientific MCP and External Data Sources

Frontend: apps/web

Agent Backend: Native Executor

Ascend NPU Host Broker Design

Single-file Binary Packaging and Releases

Content-addressed storage (CAS)

Dynamic context assembly

Context assembly examples

Deployment Runtime Internals

Deep Developer Guide

Evolution sidecar: architecture, engines, and standalone deployment

Idea Tree autonomous-research engine

MCP Backend Design

Network Proxy Mechanism

PDF Extraction: services/paper

Plugin mechanism

External Data-Source Rate Limiting

Repository Layout Reference

Review and Provenance

Project and Session Runner inheritance

Runtime Core boundaries

Session trajectory and model context

Skill Library: Current Implementation

Skill Self-Evolution: Current Implementation