Orkestra SDK Overview - Orkestra

Welcome to the Orkestra Python SDK. Build multi-agent workflows, chain logic with handlers, get structured outputs with Pydantic, and deploy workflows as API endpoints.

Installation

pip install orkestra-sdk  # or use poetry

Create a client and an agent

from orkestra import Orkestra, LLMProvider

client = Orkestra()
agent = client.Agent(
  name="Summarizer",
  description="Summarizes text",
  model_provider="openai",
  model_name="gpt-3.5-turbo",
  api_secret="YOUR_OPENAI_API_KEY",
)
print(agent.generate("Summarize Orkestra in one sentence."))

Build a workflow

researcher = client.Agent(
  name="Researcher",
  description="Gathers facts",
  model_provider="openai",
  model_name="gpt-4-turbo",
  api_secret="YOUR_OPENAI_API_KEY",
)

summarizer = client.Agent(
  name="Summarizer",
  description="Summarizes findings",
  model_provider="openai",
  model_name="gpt-3.5-turbo",
  api_secret="YOUR_OPENAI_API_KEY",
)

wf = client.Workflow().add(researcher).add(summarizer)
print(wf.run("Explain Retrieval-Augmented Generation in simple terms"))

Deploy as an API (FastAPI + Swagger)

from orkestra import OrkestraServer
server = OrkestraServer(client)
server.add_workflow(
  endpoint_name="summarize",
  workflow=wf,
  summary="Summarize topic",
  description="Researches a topic then summarizes the findings.",
)
server.run()  # Visit /docs or /redoc

Continue with: Agents, Workflows, Structured outputs, Conditional workflows, and Server.