VexDB Dify Plugin

Provides one-stop integration capabilities for the VexDB Vector Database within Dify Applications, covering database browsing, table management, data writing, and vector similarity retrieval. It is suitable for scenarios such as RAG, semantic search, and Embedding storage and retrieval.

---

Open Source Note

This repository provides the complete implementation of the VexDB Dify Plugin, suitable for:

• Users wishing to deploy Dify + VexDB in local/private environments.
• Developers wishing to perform secondary development based on existing capabilities.
• Teams wishing to integrate VexDB vector database capabilities into Agents / Workflows.

---

Capabilities Overview

Capability	Identity (ID)	Description
Discover Databases	list_databases	List all non-template databases in the VexDB instance.
Manage Tables	list_tables	List all tables in a schema, with support for filtering to return only tables containing vector columns.
Schema Awareness	table_schema	Retrieve metadata such as column names, data types, and indexes.
Data Writing	insert_row	Insert a single row of data, supporting JSON format and floatvector types.
Vector Search	vector_search	Perform TopK similarity search based on a floatvector column.

---

Detailed Usage Guide (Input Format Specifications)

When orchestrating in a Dify Agent or Workflow, please strictly adhere to the following parameter input specifications, especially regarding JSON data structures.

1. Insert Row

Insert a single row into the specified table.

• Table Name (table): The target table name (String).
• Row Data (row): ⚠️ Must be a valid JSON object string.
  • The plugin internally uses a JSON parser to process this field, so property names must use double quotes.
  • Vector data must be represented as an array of floating-point numbers.
  • LLM Input Example:

    {
      "content": "This is a test document fragment",
      "metadata": {"source": "wiki", "author": "admin"},
      "embedding": [0.012, -0.34, 0.88, 0.15]
    }

• Schema: Defaults to public.

2. Vector Search

Perform a similarity search.

• Table Name (table): The target table name (String).
• Vector Column (vector_column): The field name storing vector data (e.g., embedding).
• Query Vector (query_vector): ⚠️ Must be a valid JSON array string.
  • Format must be [float, float, ...].
  • LLM Input Example:

    [0.012, -0.34, 0.88, 0.15]

• Distance Type (distance_type): Optional. Values: l2 (Default, Euclidean), cosine (Cosine), ip (Inner Product).
• Top K: Number of results to return. Defaults to 5.

3. Get Table Schema

• Table (tables):
  • Empty: Returns the schema for all tables in the current schema.
  • Single Table: Enter the table name, e.g., users.
  • Multiple Tables: Use commas to separate, e.g., users, items.

4. List Tables

• Only Vector Tables (only_vector_tables): Boolean (true/false). If set to true, the plugin automatically filters out tables that do not contain fields with the vector type.

5. List Databases

• Lists all non-template databases in the VexDB instance.
• DB URI (db_uri): Optional database connection string.

---

Design Philosophy

• Weak Assumption: Does not strictly depend on specific table structures, making it suitable for multiple business scenarios.
• Plug and Play: Authorize once to use; supports db_uri override when necessary.

---

Security & Connection

• Global Configuration: Database connection information is usually configured globally via Dify's Secret Input when installing the plugin.
• Temporary Override: All tools provide a db_uri parameter. If this parameter is filled when calling a tool, it will override the global configuration (useful for debugging or multi-database operations).
• Privacy & Safety: The plugin code captures exceptions and does not throw raw stack trace information in the Tool response, avoiding the leakage of sensitive connection strings.

---

📄 License

Maintained and provided by shuzhiyinhang.

Forking, modifying, and redistributing are welcome under the terms of the License.