Keyboard shortcuts

Press ← or → to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

Yanshi Manual

Looking for the interactive overview, visual routing gallery, and 15-template matrix? Visit the Yanshi Product Home.

Yanshi is a multimodal content-production runtime written in Rust. Sources go in — an idea, a URL, uploaded files, a recording — and finished deliverables come out: reports, mind maps, audio overviews, slides, flashcards, quizzes, infographics, data tables, video overviews and domain engineering diagrams. Every deliverable runs through the same five stages (Sources → Planner → Generators → Renderers → Artifacts), described in RFC: Content Studio Runtime. It is not a coding-agent framework.

The runtime underneath offers two complementary execution styles that share one foundation:

  • Deterministic DAG workflows — a Flow of nodes with explicit inputs and outputs, dependency-ready concurrent scheduling, conditional execution, map/while loops, retry, timeout, and checkpointed recovery. Usually written as YAML and run with yanshi workflow run.
  • Agent-native loops — ReActAgent, Plan-Execute, and multi-agent supervisors that plan, call tools, reflect, and keep memory, with structured step/event traces and hard runtime limits.

The two compose in both directions: AgentNode embeds an agent as a step of a DAG, and WorkflowTool exposes a whole DAG as a tool an agent can call. Around them sit Skills (declarative capability packages), MCP tools, RAG, memory, tracing, the Harness governance shell (hooks, approvals, audit), and an HTTP gateway with a Web UI.

How this manual is organized

PartRead it when you want to…
Content Studio Catalogue & Desktoprun the 15 built-in slide, video, paper, repo, and report generators or install Yanshi Studio desktop
Getting Startedconfigure providers and run a first workflow without API keys
Workflowswrite, debug, and make DAG workflows resilient
Agentsrun agent loops, multi-agent setups, and Harness sessions
Skills, Tools and Pluginspackage capabilities and control what tools may do
MCPconsume or serve Model Context Protocol tools
Models, Multimodal and RAGpick a provider/model and use image, audio, or retrieval
Observabilitycapture, replay, and inspect traces
Operationsdeploy the gateway, workers, and database
Extending Yanshiwrite custom runtimes, nodes, or strategies
Referencecheck stability tiers, compatibility, and release notes
Design and Internalsunderstand the crate architecture, the Content Studio RFC, the evaluations and the other RFCs behind it

API documentation for individual crates is generated by cargo doc; links into source files from this manual open the file on GitHub.

The source of every page lives in the repository’s docs/ directory — use the edit icon at the top of a page to propose a change.