Skip to content

OpenAI Client

Create a model

from openai import OpenAI
from gimkit import from_openai

client = OpenAI()  # reads OPENAI_API_KEY
model = from_openai(client, model_name="gpt-4o")

Prompt recommendation

For OpenAI paths, prefer use_gim_prompt=True.

Use natural language with masked tags directly in an f-string:

from gimkit import guide as g

query = f"""
Name: {g.person_name(name="name")}
Email: {g.e_mail(name="email")}
"""

result = model(query, output_type="json", use_gim_prompt=True)

If your OpenAI provider does not support JSON constraints, fall back to output_type=None:

result = model(query, output_type=None, use_gim_prompt=True)

Output types

`output_type="json"` (recommended)

Use JSON schema constraints and convert JSON fields back into tag results.

result = model(query, output_type="json", use_gim_prompt=True)
print(result.tags["name"].content)

`output_type=None` (fallback)

Use this when your OpenAI provider does not support JSON constrained output.

result = model(query, output_type=None, use_gim_prompt=True)
print(result.tags["email"].content)

Advanced options

include_grammar=True: inject grammar text into the query string before generation.
backend: pass through to Outlines generator backend selection.
**inference_kwargs: forwarded to the underlying OpenAI call.