AI Agent Management
Memory Database
Use Memory Tables to give your AI agent persistent, structured memory — store user preferences, conversation history, lead data, and more across sessions.
Overview
Memory Database allows your AI agent to remember information across conversations. Unlike ephemeral chat context, memory tables provide persistent, structured storage that your workflows can read from and write to. Each workspace can have multiple memory tables, each with a custom schema you define.
Creating Memory Tables
Navigate to the Memory Database section in your workspace sidebar. Click "Create Table" to define a new table. Give it a name and description, then define columns with types such as string, number, boolean, date, json, or text. Each column can be marked as required, unique, or searchable. System columns (id, createdAt, updatedAt) are added automatically.
Schema Management
You can update your table schema at any time — add new columns, rename existing ones, change types, or toggle constraints like required and searchable. Columns marked as "searchable" are indexed for semantic vector search, enabling natural-language queries against your stored data.
Working with Rows
Insert rows manually through the UI or programmatically via workflow cards. You can query rows with filters (equals, contains, greater than, etc.), combine conditions with AND/OR logic, and sort or paginate results. Bulk insert is supported for importing large datasets.
Semantic Search
When columns are marked as searchable, Docimal generates vector embeddings for their content. Use the Search Memory card in workflows to perform natural-language queries that return the most relevant rows ranked by similarity score.
Memory Cards in Workflows
Six dedicated workflow cards interact with memory: Insert Memory (add a new row), Update Memory (modify a row by ID), Delete Memory (remove a row), Filter Memory (natural-language to structured query), Search Memory (semantic vector search), and Query Memory (retrieve a row by ID). Use these cards in your workflows to build stateful, context-aware conversations.
Best Practices
Keep table schemas focused — one table per domain concept (e.g., leads, preferences, orders). Mark only the columns you actually need to search as "searchable" to optimize performance. Use the Filter Memory card for exact lookups and Search Memory for fuzzy, natural-language queries. Regularly review stored data to ensure quality.