Skip to main content

Intent Classification

One of Synentra's core capabilities is Intent Classification.

Instead of making decisions solely based on HTTP methods, URLs, or identities, Synentra analyzes requests to determine what an AI agent intends to accomplish.

Understanding intent enables policies to be written in business language rather than low-level API details.


What Is Intent?

Intent describes the goal behind a request.

Consider these requests:

DELETE /users/123
DELETE /sessions/123

Both use the same HTTP method.

However, they represent very different operations.

One deletes customer data.

The other removes a temporary session.

Traditional API gateways treat both requests similarly.

Synentra recognizes that their intents are different.


Why Intent Matters

HTTP requests often fail to communicate business meaning.

For example:

POST /api/export

could mean:

  • Export one document
  • Export one customer
  • Export one million records

The endpoint alone provides little context.

Intent classification provides the missing semantic understanding.


Semantic Analysis

Synentra analyzes multiple aspects of each request, including:

  • HTTP method
  • URL path
  • Query parameters
  • Headers
  • Request body
  • Contextual metadata

These features are evaluated by a local machine learning model to infer the most likely intent.


Example

Request:

POST /employees/export

Payload:

{
"department": "Finance",
"format": "csv"
}

Predicted intent:

bulk_export

Confidence:

97%

Policies can then be written against the semantic action rather than the endpoint.


Local Inference

Synentra performs inference locally using an ONNX Runtime model.

Advantages include:

  • Low latency
  • No external API calls
  • Improved privacy
  • Offline operation
  • Predictable performance

Because requests never leave your environment, sensitive enterprise data remains under your control.

Was this helpful?