Dify Knowledge Plugin

Tool plugin for the Dify Knowledge API. It lets you inspect knowledge bases, inspect documents, retrieve chunks, and query the available models used for knowledge-base configuration.

Features

• list_datasets: list knowledge bases with optional keyword, pagination, permission, and tag filters.
• list_documents: list documents in a knowledge base with optional keyword, status, and pagination filters.
• retrieve_chunks: retrieve normalized chunk results from a knowledge base.
• list_available_models: query available Dify models by type, especially text-embedding and rerank for knowledge-base setup.

Credentials

• Base URL: Dify Knowledge API base URL. Both https://your-dify-host/v1 and https://your-dify-host are accepted. The plugin normalizes the latter to /v1.
• API Key: Dify Service API key with access to the target workspace knowledge resources.

Getting Credentials

Open the Service API panel in Dify, then copy the API endpoint into the plugin Base URL field.

Click API Key in the Service API panel, then copy the key from the API key dialog into the plugin API Key field.

Tool Reference

list_datasets

Used to inspect available knowledge bases before retrieval or document lookup.

• keyword: optional name filter.
• page: optional page number, defaults to 1.
• limit: optional page size, defaults to 20.
• include_all: optional flag to include all visible knowledge bases regardless of permission scope.
• tag_ids: optional comma-separated list or JSON array string of tag ids.

Getting dataset_id

Open a knowledge base in Dify and copy the UUID between /datasets/ and /documents from the browser address bar.

list_documents

Used to inspect documents inside a specific knowledge base.

• dataset_id: required knowledge base id.
• keyword: optional document-name filter.
• status: optional status filter.
• page: optional page number, defaults to 1.
• limit: optional page size, defaults to 20.

retrieve_chunks

Used to retrieve relevant chunks from a specific knowledge base.

Optional retrieval parameters only affect the current request and are not saved to the knowledge base. If an optional retrieval parameter is left empty, Dify uses the retrieval configuration already saved in that knowledge base.

• dataset_id: required knowledge base id.
• query: required search query.
• search_method: optional retrieval-method override for internal knowledge bases.
• top_k: optional result-count override.
• score_threshold_enabled and score_threshold: optional similarity-threshold overrides.
• reranking_enable and reranking_mode: optional reranking overrides for internal knowledge bases.
• reranking_provider_name and reranking_model_name: required when reranking_mode is reranking_model. Available values can be obtained from list_available_models with model_type=rerank.
• weight_type, vector_weight, and keyword_weight: optional hybrid-search weight overrides.
• attachment_ids: optional comma-separated list or JSON array string of attachment ids.

list_available_models

Wraps Dify's "Get Available Models" endpoint: GET /workspaces/current/models/model-types/{model_type}.

• model_type: required model type. Supported values are text-embedding, rerank, llm, tts, speech2text, and moderation.
• Output: normalized provider groups with provider_count, total model_count, and each provider's models list.

Notes

• The model-list tool is mainly useful when you need to know which text-embedding or rerank values can be used in knowledge-base configuration.
• The retrieval tool only sends retrieval-model overrides when you explicitly provide them. These overrides are request-scoped and do not change the saved knowledge-base configuration.
• CSV strings like a,b,c and JSON array strings like ["a","b","c"] are both accepted for list-style parameters such as tag_ids and attachment_ids.

Support

If this plugin is helpful, a GitHub Star is appreciated. For bugs, questions, or feature requests, please open an Issue with your Dify version, plugin version, and the steps to reproduce the problem.