Getting Started

Client Setup

Set up Instructor with different LLM providers. Each provider requires slightly different setup.

import instructor
from openai import OpenAI

Default mode for OpenAI is TOOLS (function calling)

client = instructor.from_openai(OpenAI())

You can also specify a different mode

client = instructor.from_openai(
    OpenAI(),
    mode=instructor.Mode.JSON  # Use JSON mode instead
)

import instructor
from anthropic import Anthropic

Default mode is TOOLS for Claude 3

client = instructor.from_anthropic(Anthropic())

Use JSON mode if needed

client = instructor.from_anthropic(
    Anthropic(),
    mode=instructor.Mode.JSON
)

import instructor
import google.generativeai as genai

Configure Gemini

genai.configure(api_key="YOUR_API_KEY")
model = genai.GenerativeModel("gemini-1.5-flash")

Gemini needs a specific mode

client = instructor.from_gemini(
    model,
    mode=instructor.Mode.GEMINI_TOOLS  # or GEMINI_JSON
)

import instructor
import cohere

Create Cohere client

cohere_client = cohere.Client("YOUR_API_KEY")

Patch with instructor

client = instructor.from_cohere(cohere_client)

import instructor
from mistralai.client import MistralClient

mistral_client = MistralClient(api_key="YOUR_API_KEY")

Patch with instructor

client = instructor.from_mistral(mistral_client)

Instructor supports different modes for different providers:

from instructor import Mode

Available modes

Mode.TOOLS         # OpenAI function calling format (default for OpenAI)
Mode.JSON          # Plain JSON generation
Mode.MD_JSON       # Markdown JSON (used by some providers)
Mode.ANTHROPIC_TOOLS # Claude tools mode (default for Anthropic)
Mode.GEMINI_TOOLS  # Gemini tools format
Mode.GEMINI_JSON   # Gemini JSON format

Running the Example

First, install Instructor and any dependencies

$ pip install instructor pydantic

Run the Python script

$ python client-setup.py

Further Information

Documentation link 1