ai-integration

AI Integration

Core APIs

from iterable.ai import doc
from iterable.ai.context import sample_for_llm, redact_for_llm
from iterable.catalog import describe_format, export_catalog, list_formats
from iterable.ops import inspect

Documentation generation

# Local provider (no API key)
doc.generate("data.csv", provider="lmstudio", base_url="http://localhost:1234/v1")

# With inspect
inspect.analyze("data.csv", autodoc=True, autodoc_provider="openai")

Install: pip install iterabledata[ai] or pip install -e ".[ai]".

Block-based documentation (generate_blocks)

from iterable.ai import doc

result = doc.generate_blocks(
    "data.csv",
    blocks=["general", "schema", "quality", "examples", "statistics"],  # codebook also available
    context={"title": "Population", "territory": "Russia"},
    progress=lambda e: print(e.stage.value, e.progress),
)
result["blocks"]["schema"]["data"]["fields"]   # structured data per block
result["full_document_markdown"]               # assembled document

• Each block returns {markdown, data}; statistics is computed (no LLM); lineage/geo_coverage are deferred stubs.
• LLM blocks use structured output (provider.generate_structured) with Pydantic models in iterable.ai.models.
• Sampling adapts to file size (MAX_ROWS_SAMPLING); stats support null_fraction, top_values, is_dictionary (DICT_THRESHOLD).
• Provider config via LLM_PROVIDER/LLM_BASE_URL/LLM_API_KEY/LLM_DEFAULT_MODEL; provider="openai-compatible" targets any OpenAI-compatible endpoint.
• doc.generate(..., blocks=[...]) delegates to generate_blocks; without blocks the legacy single-document path is unchanged.

Safe LLM sampling

Always sample and redact before cloud APIs:

rows = sample_for_llm("data.csv", max_rows=10, strategy="stratified")
safe = redact_for_llm(rows)

Format catalog for agents

describe_format("xml")  # includes example_args, limitations, capabilities
export_catalog(format="json")  # full catalog; committed at dev/formats.json