Memory Banks

Memory Banks are Intentgine’s mechanism for providing persistent context to the resolution engine. They act as “long-term memory” for your AI, allowing it to learn from corrections and adapt to your specific domain vocabulary.

What is a Memory Bank?

A Memory Bank is a collection of items. Each item maps a natural language query to a specific tool call.

{
  "query": "Reset password for user 123",
  "correct_tool": "admin_reset_password",
  "correct_params": { "user_id": 123 }
}

When you import this item into a Memory Bank, Intentgine indexes the query. During a /v1/resolve call, if the user’s input is semantically similar to this query, the resolution engine retrieves this example and provides it as context.

Why use Memory Banks?

1. Domain Adaptation: Teach the system your specific jargon (e.g., “nuke it” -> delete_database).
2. Correction: Fix edge cases where resolution consistently fails.
3. Consistency: Ensure similar queries yield identical tool structures.
4. Portability: Assign the same Memory Bank to multiple Apps (e.g., share a “Core Commands” bank across Dev and Prod).

How it Works

1. Creation: Create a bank via the Console or API.
2. Assignment: Assign the bank to an App. You can assign multiple banks with different priorities.
3. Resolution: When a request comes in:
   • The system searches all assigned banks for relevant items.
   • The top matching items are injected as context.
   • The system uses these examples to guide its decision.

Managing Banks

• Import: Bulk import thousands of examples via CSV/JSON.
• Correct: Use the /v1/correct endpoint to programmatically add new items based on user feedback.
• Visibility: Banks can be private (only your team) or public (shared with the community - coming soon).