documind 0.2.2

DocuMind SDK - lightweight AI document generation for PPTX, DOCX, Markdown, XLSX, and HWPX

DocuMind SDK

Korean: README.ko.md

Lightweight Python SDK for AI document generation.

Install documind when you want to generate native document files directly from Python.

Installation

pip install documind

Provider integration dependencies for OpenAI, Bedrock, Anthropic, Gemini, and Vertex AI are included in the default installation.

Quick Start

import asyncio
from documind import generate_document, init

init(
    llm_provider="openai",
    openai_api_key="sk-...",
    default_llm_model="gpt-4o",
    storage_local_path="./outputs",
)

async def main():
    result = await generate_document(
        query="Create an AI document automation proposal.",
        document_type="pptx",
        locale="en",
    )
    print(result.output_path)

asyncio.run(main())

SDK API Reference

Public imports:

from documind import (
    DocuMind,
    GenerationRequest,
    GenerationResult,
    ImageAttachment,
    TemplateInput,
    configure,
    generate_document,
    init,
    stream_document,
)

Engine Configuration

Use init(**config) for process-wide defaults, or pass the same values to DocuMind(...) for one engine instance.

Option	Type / Example	Description
llm_provider	"openai"	Provider route. Supported values include openai, anthropic, azure, bedrock, gcp_vertex, gemini, ollama, vllm, custom.
use_default_models	True	If true, agents use the default model names below.
default_llm_model	"gpt-4o"	Text generation model.
default_vlm_model	"gpt-4o"	Accepted for vision-model compatibility.
default_image_model	"dall-e-3"	Image asset model. If unavailable, generation continues without image assets.
storage_local_path	"./outputs"	Output directory for generated documents, HTML previews, and image assets.
log_level / log_file	"INFO" / path	Logging controls.
openai_api_key	"sk-..."	OpenAI credential.
openai_base_url	URL	OpenAI-compatible endpoint override.
anthropic_api_key	string	Anthropic credential.
google_api_key	string	Gemini API key.
gcp_project_id, gcp_location, google_application_credentials	strings	Vertex AI settings.
aws_profile, aws_region, aws_access_key_id, aws_secret_access_key, aws_session_token, aws_role_arn	strings	Bedrock settings.
azure_openai_api_key, azure_openai_endpoint, azure_openai_api_version, azure_openai_deployment	strings	Azure OpenAI settings.
custom_llm_base_url, custom_llm_api_key, custom_llm_model_name	strings	OpenAI-compatible custom provider settings.
preload_icons	False	Optional icon asset warm-up.
icon_preload_limit	integer	Optional icon warm-up limit.

Document Generation

generate_document(...) is the simplest one-shot helper. For repeated calls, create DocuMind(...) once and call engine.generate(...).

from documind import DocuMind, ImageAttachment, TemplateInput

engine = DocuMind(
    llm_provider="openai",
    openai_api_key="sk-...",
    storage_local_path="./outputs",
)

result = await engine.generate(
    query="Create a customer onboarding deck.",
    format="pptx",
    locale="en",
    template=TemplateInput(path="./template.pptx"),
    images=[
        ImageAttachment(path="./product.png", description="Product screenshot"),
    ],
    needs_research=False,
)

Argument	Type / Default	Description
query	required string	Natural-language document goal.
format	"pptx"	Output format for engine.generate. One of pptx, docx, xlsx, md, hwp.
document_type	optional	Alias for format; useful when adapting external input models.
template_id	optional string	Compatibility hook for template identifiers.
template	path, bytes, dict, TemplateInput	Optional native/template file.
images	list of paths, bytes, dicts, ImageAttachment	Optional image evidence for planning and visual references.
session_id	optional string	Caller-provided correlation/session ID.
locale	"ko"	Locale hint. Output language is also inferred from query.
needs_research	None	True forces research, False skips it, None lets DocuMind infer intent.
preload_icons	optional bool	Per-call icon warm-up toggle.
**options	dict	Forward-compatible pipeline options. Stable SDK callers should prefer the explicit arguments above.

Structured Input

from documind import GenerationRequest

generation_input = GenerationRequest(
    query="Create a weekly report.",
    document_type="docx",
    locale="en",
    needs_research=False,
    options={"template_id": "internal-template-id"},
)

result = await engine.generate_from_request(generation_input)

GenerationRequest field	Type / Default	Description
query	required string	Natural-language document goal.
document_type	"pptx"	pptx, docx, xlsx, md, or hwp.
template	None	Path, bytes, dict, or TemplateInput.
images	[]	Image evidence list.
session_id	None	Correlation/session ID.
locale	"ko"	Locale hint.
needs_research	None	Research routing toggle.
stream	False	Streaming marker for external adapters. Use generate_stream for SDK streaming.
options	{}	Additional pipeline options, including template_id.

Helper type	Fields
TemplateInput	path, content, filename
ImageAttachment	path, content, filename, mime_type, role="content_reference", description

GenerationResult exposes success, output_path, output_bytes, document_type, mime_type, fidelity_scores, slide_count, errors, metadata, and to_dict().

Supported Formats

Format	Output
pptx	Native PowerPoint file
docx	Native Word file
xlsx	Native Excel workbook
md	Markdown document
hwp	HWPX document

Streaming

from documind import DocuMind, GenerationRequest

engine = DocuMind(llm_provider="openai", openai_api_key="sk-...")
generation_input = GenerationRequest(query="Create a weekly report.", document_type="docx")

async for event in engine.generate_stream(generation_input):
    print(event.to_sse())

Build From Source

cd packages/documind
python scripts/sync_runtime.py
python -m build

Publishing instructions are in PUBLISHING.md and PUBLISHING.ko.md.

License

Apache-2.0