Dynamic Table
The Dynamic Table plugin gives your agent access to a managed, searchable data store. You define tables with typed columns, populate them with rows, and the agent can query that data using natural language — returning structured results directly in conversation.
Unlike static document uploads, data in a Dynamic Table can be added, edited, deleted, and bulk-imported at any time without redeploying or re-indexing the agent.
How It Works
When a user asks a question, the plugin:
- Receives the natural language query from the AI.
- Converts the query into a vector embedding.
- Runs a semantic similarity search against all rows in the configured table.
- Applies threshold and tolerance settings to control result quality.
- Optionally reranks results using an LLM for improved accuracy.
- Returns the matching rows as structured data for the AI to use in its response.
Key Features
| Feature | Description |
|---|---|
| Flexible schema | Define typed columns per table: text, number, email, phone, date, boolean. |
| Semantic search | Rows are indexed as vector embeddings and retrieved by semantic similarity. |
| Tolerance increment | Automatically lowers the search threshold if not enough results are found. |
| LLM reranking | Optionally rerank search results using an AI model for better relevance. |
| Primary keys | Mark one column as a primary key to enforce uniqueness per table. |
| Search weights | Assign higher importance to specific columns in the search index. |
| Import and export | Load rows from CSV or Excel files; export table data at any time. |
| Integration API | External systems can read and write rows using per-table API keys. |
Setup Flow
Follow these steps in order to configure the plugin:
- Create a table — define columns, types, and search settings.
- Add and manage rows — populate the table and keep data up to date.
- Import and export data — bulk load from CSV or Excel, or export the full table.
- Configure search settings — tune similarity threshold, tolerance increment, and reranking.
- Configure function tool settings — define how the plugin appears and behaves for the AI.
- Test semantic search — verify search quality before publishing.
- Set up the Integration API — allow external systems to read and write rows programmatically.
Prerequisites
Before creating a plugin instance:
- Know what data the table will hold and what users will search for.
- Decide which columns should be searchable (included in the vector index).
- Determine whether any column needs to be a primary key.
- Have at least a small sample of real data ready to test search quality.