llama-index-readers-feishu 0.1.0

LlamaIndex data loaders for Feishu/Lark wiki, docs, and drive

LlamaIndex Feishu Readers

Load documents from Feishu/Lark into LlamaIndex. This package provides three readers backed by the official lark-oapi SDK:

• FeishuWikiReader — load documents from wiki spaces (with recursive subnode support)
• FeishuDocsReader — load individual docx documents by token
• FeishuDriveReader — load docx documents from cloud drive folders (with recursive subfolder support)

[!NOTE] The official llama-index-readers-feishu-wiki package has been removed from LlamaIndex, and llama-index-readers-feishu-docs is no longer maintained. This project carries the work forward with expanded Feishu document loading for wiki, docs, and drive.

Installation

pip install llama-index-readers-feishu

Prerequisites

1. Create a Feishu custom app in the developer console.
2. Enable permissions based on the reader you use:

Reader	Required scopes
Wiki	wiki:wiki:readonly or wiki:node:retrieve, docs:document.content:read
Docs	docs:document.content:read or docx:document:readonly
Drive	drive:drive:readonly or drive:drive, docs:document.content:read

3. Grant the app access to the target wiki space or drive folder.

Quick Start

Wiki

from llama_index.readers.feishu import FeishuWikiReader

reader = FeishuWikiReader(app_id="cli_xxx", app_secret="your_app_secret")
documents = reader.load_data(
    space_id="your_space_id",
    parent_node_token="optional_parent_node_token",
    recursive=True,
)

Docs

from llama_index.readers.feishu import FeishuDocsReader

reader = FeishuDocsReader(app_id="cli_xxx", app_secret="your_app_secret")
documents = reader.load_data(document_ids=["docx_token_1", "docx_token_2"])

Drive

from llama_index.readers.feishu import FeishuDriveReader

reader = FeishuDriveReader(app_id="cli_xxx", app_secret="your_app_secret")
documents = reader.load_data(
    folder_token="your_folder_token",
    recursive=True,
)

The drive reader loads:

• Files with type=docx
• Shortcuts where shortcut_info.target_type=docx

Other file types (folders, sheets, bitables, etc.) are skipped. Folders are traversed when recursive=True.

Post-process Exported Text

Feishu markdown exports may embed HTML fragments such as tables. Pass plugins to normalize loaded text:

from llama_index.readers.feishu import (
    FeishuDriveReader,
    RegexPlugin,
    html_table_to_markdown_plugin,
)

reader = FeishuDriveReader(
    app_id="cli_xxx",
    app_secret="your_app_secret",
    plugins=[
        html_table_to_markdown_plugin,
        RegexPlugin(pattern=r"\n{3,}", repl="\n\n", name="collapse_blank_lines"),
    ],
)
documents = reader.load_data(folder_token="your_folder_token")

Use default_markdown_plugins() for the built-in HTML table converter and Feishu markdown unescape plugin.

User Access Token

To access content with user identity instead of tenant identity:

reader = FeishuWikiReader(
    app_id="cli_xxx",
    app_secret="your_app_secret",
    user_access_token="u-xxxxxxxx",
)

Lark (Overseas)

reader = FeishuWikiReader(
    app_id="cli_xxx",
    app_secret="your_app_secret",
    domain="https://open.larksuite.com",
)

Or switch after initialization:

reader = FeishuWikiReader(app_id="cli_xxx", app_secret="your_app_secret")
reader.set_lark_domain()

Use with LlamaIndex

from llama_index.core import VectorStoreIndex

index = VectorStoreIndex.from_documents(documents)
query_engine = index.as_query_engine()
response = query_engine.query("Summarize the content")
print(response)

Examples

Integration scripts under examples/ load live content and export markdown files. Copy .env.example to .env and fill in your credentials:

cp .env.example .env
# edit .env with your Feishu app credentials and target IDs

# Wiki
python examples/load_wiki_documents.py

# Docs
python examples/load_docs_documents.py

# Drive
python examples/load_drive_documents.py

Development

uv sync
uv run poe hooks-install
uv run poe lint
uv run poe test
uv run poe test-cov        # optional: with coverage report

License

MIT