Agently Docs

EN

中文

EN

中文

Appearance

Workflow and Extensions Overview ​

This chapter starts when your application touches the outside world, or when the control flow itself becomes more complex.

TIP

From this chapter onward, the recommended default is Async First. Tool streaming, instant structured output, FastAPI service exposure, and TriggerFlow runtime_stream plus async_emit(...) all fit async interfaces better.

When to read this ​

  • The model needs tools, search, browser access, commands, or an MCP server
  • You want to expose Agently as a web service
  • You need a knowledge base or RAG
  • You now have explicit stages, branching, concurrency, or wait/resume logic

What you will learn ​

  • Which problems belong to extensions and which already belong to workflow orchestration
  • What Tools, MCP, KB, FastAPI, and TriggerFlow each solve
  • How to upgrade from the request side without over-designing too early

Selection map ​

Reading by problem type ​

Tools ​

Good fit:

  • search, browser, command execution
  • tool logs and tool-loop control

Entry:

MCP ​

Good fit:

  • connecting an existing MCP server directly into Agently

Entry:

Knowledge Base / RAG ​

Good fit:

  • document retrieval, knowledge QA, source-backed answers

Entry:

FastAPI ​

Good fit:

  • exposing HTTP, SSE, or WebSocket APIs

Entry:

TriggerFlow ​

Good fit:

  • multi-stage execution
  • branches and concurrency
  • runtime stream
  • wait, resume, and execution save/load
  • consuming model results inside async chunks and routing them further

Entry:

Common mistakes ​

  • Building a whole workflow when all you needed was one external API call
  • Mixing retrieval output directly into prompts and business logic before clarifying ownership
  • Reinventing a streaming protocol before checking FastAPIHelper and TriggerFlow runtime stream

Next ​

  • agently-agent-extensions
  • agently-knowledge-base
  • agently-triggerflow