Nimble overview

Nimble is a family of models from Bespoke Labs for text classification and other judgments about text or JSON. Nimble answers with probabilities, so you can set your own thresholds. You call Nimble at POST https://api.bespokelabs.ai/v1/systemone. The model field in the request picks the model.

​

What you can build

Use Nimble wherever your code needs a judgment about text: classifying it, routing it, checking it against a rule, or rating it. Nimble works zero-shot. You describe each option in plain language, and there’s no classifier to train.

​

Models

Every model answers up to 64 typed questions about a piece of text or JSON. nimble-latest currently points to nimble-v4-max, and it will move to newer models as we release them. To stay on one model, request it by name, e.g., nimble-v4-max. Different models return different probabilities for the same request. If you tuned thresholds on the answers, check them again when you switch models or when nimble-latest moves to a new one.

​

The request format

Nimble takes one request format, which is called System One. A request has three fields:

• model names the model.
• state is the text or JSON to ask about.
• questions maps your own question IDs to questions. The answers use the same IDs.

You can send the requests with plain HTTP or with TypeSafe’s Python SDK. The path /v1/nimble/systemone takes the same requests as /v1/systemone. Nimble has three types of question:

• A yes/no question (noul) asks whether a statement is true. The answer is the probability that it is true.
• A multiple-choice question (choice) asks which of 2 to 255 options fits best. The answer names the most likely option and gives the probability of every option.
• A rating question (score) asks where the state falls on a scale of 2 to 255 ordered levels. The answer is the expected level.