StructGuard

Reliable, provider-agnostic structured outputs for production LLM applications.

StructGuard makes sure every LLM response conforms to a schema you define — Pydantic, dataclass, TypedDict, or raw JSON Schema — before it reaches your application code. It handles prompting, JSON repair, validation, and intelligent retries in one call, across OpenAI, Anthropic, Gemini, Groq, and Ollama.

Install

pip install structguard              # core only
pip install structguard[openai]      # + OpenAI SDK
pip install structguard[anthropic]   # + Anthropic SDK
pip install structguard[all]         # every provider SDK

Quick start

from pydantic import BaseModel
from openai import OpenAI
from structguard import generate

class Incident(BaseModel):
    summary: str
    priority: str
    confidence: float

client = OpenAI()

incident = generate(
    llm=client,
    prompt="Summarize this incident: Database connection pool exhausted, causing 500s.",
    schema=Incident,
)

print(incident.summary, incident.priority, incident.confidence)

With a full report

result = generate(llm=client, prompt=prompt, schema=Incident, return_report=True)
print(result.data)
print(result.report)   # ✓ Success | provider=OpenAI model=gpt-4o-mini retries=0 ...

Error handling

from structguard import RetryLimitExceeded, SchemaValidationError

try:
    incident = generate(llm=client, prompt=prompt, schema=Incident, max_retries=2)
except RetryLimitExceeded as e:
    print(f"Gave up after {e.attempts} attempts: {e.last_error}")

Supported schema types

• pydantic.BaseModel
• stdlib @dataclass
• TypedDict
• raw JSON Schema dict

Supported providers

OpenAI · Anthropic · Google Gemini · Groq · Ollama

Development

git clone https://github.com/sujanrupu/structguard
cd structguard
pip install -e ".[dev]"
pytest

License

MIT