slimschema 0.0.0.dev12

Token-efficient schema language for LLMs with validation and conversion

SlimSchema

The bridge between LLMs, Python types, and fast validation.

SlimSchema is a schema language optimized for LLM token efficiency (70% smaller than JSON Schema) that converts seamlessly between YAML, Pydantic, and msgspec.

Core Features

• Token Efficient: Compressed syntax (age: 18..120) saves context window and costs.
• Deep Validation: Recursive validation for nested objects and dictionaries.
• Robust Extraction: extract JSON/XML from noisy LLM markdown.
• Universal Bridge: Convert any format to any format (YAML ↔ Pydantic ↔ msgspec).

Quick Start

pip install slimschema

The Magic Example

Define a schema with nested types and defaults, extract data from an LLM response, and get a validated Python object.

from slimschema import to_data, to_pydantic

# 1. Define a compact schema (with nested defaults!)
schema = """
user:
  name: str{3..50}
  role: admin | user | guest = "guest"
  settings:
    theme: str = "dark"
    notifications: bool = true
"""

# 2. Validate LLM output (even if partial)
# Input missing 'settings' -> defaults apply automatically
llm_output = '{"user": {"name": "Alice"}}'
data, error = to_data(llm_output, schema)

print(data)
# {
#   "user": {
#     "name": "Alice",
#     "role": "guest",
#     "settings": {"theme": "dark", "notifications": True}
#   }
# }

# 3. Convert to Pydantic for your app
UserModel = to_pydantic(schema)
instance = UserModel(**data)
print(instance.user.settings.theme) # "dark"

Documentation

Guides

• Syntax Guide: Primitives, constraints, and inline objects.
• Validation Guide: How to_data validates and reports errors.
• Defaults Guide: Handling partial data with scalar and nested defaults.
• LLM Workflow: Generating prompts and robust extraction.
• Inference Guide: Bootstrapping schemas from data.

Integrations

• Pydantic: Round-trip conversion and nested models.
• msgspec: High-performance validation structs.

Advanced

• JSON Patch: RFC 6902 support.

API Reference

• to_data(response, schema): Extract and validate JSON.
• to_schema(obj): Convert Pydantic/msgspec/YAML to SlimSchema IR.
• to_prompt(schema): Generate token-optimized prompts.
• from_data(data): Infer schema from examples.
• to_pydantic(schema): Generate Pydantic models.
• to_msgspec(schema): Generate msgspec structs.

License

MIT