Knowledge Control - Datasets

knowledge-control-datasets

Knowledge-control-datasets is the dataset management tool plugin in the knowledge-control series. It exposes the Dify “数据集/知识库” API as a set of small, composable tools: list, create, update, delete, inspect details, and test retrieval. Each action is a separate tool to keep user prompts simple and predictable.

• Author: StarsWhere
• Version: 0.0.1
• Type: Dify Tool Plugin (Python 3.12)
• License: Apache-2.0
• Repository: https://github.com/StarsWhere/knowledge-control-datasets
• Contact: starswherevip@gmail.com

What it does

• Manage datasets in a workspace: create, update, delete, view detail, list all.
• Test retrieval against a dataset with optional rerank settings.

Tools (one action per tool)

• datasets-list: List datasets with keyword/tag filter and pagination.
• datasets-create: Create a dataset; supports permission, indexing, embedding model, retrieval/rerank settings, summary index.
• dataset-detail: Get dataset detail by ID.
• dataset-delete: Delete a dataset by ID.
• dataset-update: Update dataset fields (same options as create).
• dataset-retrieve: Test search/retrieve chunks from a dataset.

Credentials (provider-level)

• api_base (optional): Defaults to https://api.dify.ai/v1. Point this to your self-hosted Dify API if needed.
• api_key (required): Workspace API Key with dataset permissions.

Key behaviors / defaults

• Embedding & rerank models: fields are optional. If left blank, Dify uses the workspace/dataset defaults.
• Rerank mode options (dropdown): reranking_model or weighted_score. Leaving it empty keeps the dataset’s existing setting.
• Retrieval options (semantic_search, full_text_search, hybrid_search, keyword_search, top_k, score_threshold, rerank_enable) are optional; unset values do not override existing dataset settings.

Quick start (local debug)

1. Install deps in your venv: pip install -r requirements.txt
2. Configure credentials in Dify when adding the plugin (api_base/api_key).
3. Run locally for debug: python -m main (or your IDE’s run command).
4. In Dify, call the tool you need; no aggregator is required.

Typical workflows

• Create a dataset: In datasets-create, fill name and optional settings; leave embedding/rerank fields empty to use workspace defaults or fill them directly if you know the model ids.
• Tune retrieval: Use dataset-update to change search_method, top_k, rerank_enable/mode, thresholds.
• Test: Use dataset-retrieve with a query to verify retrieval quality before wiring into an app.

Known limitations

• Dify tool forms are static; embedding/rerank fields cannot be auto-populated—enter model names manually or leave blank for defaults.
• Network access is required from the plugin runtime to reach your api_base.

Repo layout (key files)

• provider/knowledge-control-datasets.yaml — provider definition & credentials schema.
• tools/*.yaml — per-tool schemas and labels.
• tools/*.py — per-tool implementations.
• tools/common.py — shared HTTP client and helpers.
• LICENSE — Apache-2.0, Copyright 2026 StarsWhere (contact: starswherevip@gmail.com).

Part of the knowledge-control family

This plugin focuses on datasets/knowledge bases. Other knowledge-control series plugins will manage adjacent capabilities; they can be installed independently.