CrewAI - Neosantara Docs

CrewAI is a production-grade multi-agent orchestration framework. It allows autonomous agents to collaborate by assuming specific roles, sharing context, and executing interdependent tasks.

Setup

Install the CrewAI package:

pip install -U crewai

export NEOSANTARA_API_KEY="nsk_your_api_key_here"

NEOSANTARA_API_KEY=nsk_your_api_key_here

$env:NEOSANTARA_API_KEY="nsk_your_api_key_here"

LLM Initialization

Use the crewai.LLM class with the neosantara/<model> provider prefix to connect Neosantara models to your CrewAI agents. Credentials can be passed explicitly or automatically loaded from the NEOSANTARA_API_KEY environment variable.

import os
from crewai import LLM

# Initialize model using neosantara/ prefix
llm = LLM(
    model="neosantara/gemini-3.8-flash",
    api_key=os.environ["NEOSANTARA_API_KEY"]
)

Multi-Agent Workflow Example

The following example orchestrates two agents: a Researcher that gathers insights, and a Writer that compiles a concise technical overview from the research.

import os
from crewai import Agent, Task, Crew, Process, LLM

llm = LLM(
    model="neosantara/gemini-3.8-flash",
    api_key=os.environ["NEOSANTARA_API_KEY"]
)

# Define Agents
researcher = Agent(
    role="AI Infrastructure Analyst",
    goal="Identify emerging trends in Indonesian LLM gateways and routing latency",
    backstory="You are an experienced systems researcher specializing in Indonesian AI infrastructure and latency optimization.",
    llm=llm,
    verbose=True
)

writer = Agent(
    role="Technical Content Strategist",
    goal="Compose a crisp 2-paragraph briefing based on the research report",
    backstory="You are a technical writer who communicates infrastructure concepts clearly and concisely.",
    llm=llm,
    verbose=True
)

# Define Tasks
task_research = Task(
    description="Analyze the key advantages of Indonesia's AI Gateway for local developer teams over US-only endpoints.",
    expected_output="A bulleted summary highlighting latency benefits, local billing currency, and Indonesian data guardrails.",
    agent=researcher
)

task_write = Task(
    description="Write a 2-paragraph technical briefing using the research output.",
    expected_output="A 2-paragraph executive briefing ready for publication.",
    agent=writer
)

# Assemble and Kick off the Crew
crew = Crew(
    agents=[researcher, writer],
    tasks=[task_research, task_write],
    process=Process.sequential,
    verbose=True
)

result = crew.kickoff()
print(result)

Next Steps