TypeSafe AI

Typesafe AI provides a platform for structured, explainable AI-powered decision-making using typed questions and contextual state.

Typesafe AI is an advanced AI decision engine designed to evaluate, score, and classify unstructured data using customizable models. This connector enables Swimlane Turbine users to automate complex decision-making processes—such as single-choice selection, yes/no probability assessments, and multi-level scoring—directly within their security workflows. By integrating Typesafe AI, security teams can leverage AI-driven evaluations to triage incidents, prioritize alerts, and enrich case data without manual intervention. The connector also allows users to retrieve available AI models and flexibly tailor automation to their unique operational needs, enhancing both accuracy and efficiency in security operations.

Limitations

Text only. No image, audio, or video. Serialize a JSON object or array to a string before binding it as state.

Rate limits (429) and overload (529) are returned to the action. The definition runner does not retry. A 529 is transient, so add an on-failure retry path when the flow cannot miss a call.

The ask_* actions always publish the answer under the question id result. Use Evaluate when you need more than one question id.

model is required. A blank Model field is omitted from the body and the API returns 422. The actions default it to jev-latest.

Choice criteria is one object with 2 to 255 options. Score criteria is an array of 2 to 10 strings. Bind Noul and Choice criteria as object expressions.

Supported Versions

This TypeSafe AI connector uses the v1 API (/v1/systemone and /v1/models).

Additional Documents

Documentation TypeSafe API

Models and aliases TypeSafe models

Configuration

Prerequisites

To use the Typesafe AI connector, ensure you have the following prerequisites:

HTTP Bearer authentication to api.typesafe.ai using these parameters:

URL: The endpoint for the Typesafe AI API (e.g., https://api.typesafe.ai)

API Key: A valid API key generated from your Typesafe AI dashboard, used for authenticating requests

Authentication Methods

URL: The endpoint URL for accessing the TypeSafe API. Leave the default unless you are pointed at a private endpoint.

API Key: A key used to authenticate API requests. Do not log this value.

Verify SSL Certificates: Verify TLS certificates on outbound requests. Default is true.

HTTP(s) Proxy: Optional proxy for outbound requests. Leave blank for none.

A 401 means the API key is missing or invalid.

Capabilities

This TypeSafe AI connector provides the following capabilities:

Evaluate

Ask Noul

Ask Choice

Ask Score

List Models

Evaluate

POST /v1/systemone. Evaluate state against a map of typed questions and return answers keyed the same way Click Here.

Ask Noul

POST /v1/systemone with one noul (yes/no probability) question stored under answers.result Click Here.

Ask Choice

POST /v1/systemone with one choice question stored under answers.result. criteria is a map of option id to rubric, with 2 to 255 options Click Here.

Ask Score

POST /v1/systemone with one score question stored under answers.result. criteria is an ordered list of 2 to 10 level descriptions Click Here.

List Models

GET /v1/models. Returns aliases and models this account may send in the model field Click Here.

Configurations

TypeSafe API Key

Authenticates to api.typesafe.ai with a Bearer API key from the TypeSafe dashboard.

Configuration Parameters

Parameter	Description	Type	Required
url	TypeSafe API base URL. Leave the default unless you are pointed at a private endpoint.	string	Required
token	TypeSafe API key. Sent as Authorization Bearer. Do not log this value.	string	Required
verify_ssl	Verify TLS certificates on outbound requests.	boolean	Optional
http_proxy	Optional HTTP(S) proxy for outbound requests. Leave blank for none.	string	Optional

Actions

Ask Choice

Send a single-choice question to Typesafe AI and receive the selected answer in answers.result. Requires state, model, and questions in the request body.

Endpoint

URL: v1/systemone

Method: POST

Input

Argument Name	Type	Required	Description
state	string	Optional	Content to evaluate. Plain text, or a JSON string such as a serialized ticket, chat log, or record.
model	string	Optional	Model id or alias. Default jev-latest.
questions	object	Optional	Parameter for Ask Choice
questions.result	object	Required	Result of the operation
questions.result.type	string	Required	Type of the resource
questions.result.instructions	string	Required	What the model should decide among the criteria options.
questions.result.criteria	object	Required	Bind an option-id to rubric object. Use null when an option needs no extra detail. At least 2 options and at most 255.

Input Example

{"state":"string","model":"jev-latest","questions":{"result":{"type":"choice","instructions":"string","criteria":{}}}}

Output

Parameter	Type	Description
status_code	integer	HTTP status code of the response
reason	string	Response reason phrase
model	string	Output field: model
answers	object	Output field: answers
answers.result	object	Result of the operation
answers.result.type	string	Type of the resource
answers.result.choice	string	Highest-probability option id.
answers.result.probabilities	object	Every option mapped to its probability. Values sum to 1.
answers.result.confidence	number	How certain the model is, from 0 to 1.
usage	object	Output field: usage
usage.input_tokens	integer	Input data for the action
usage.output_tokens	integer	Output field: usage.output_tokens

Output Example

{"model":"string","answers":{"result":{"type":"string","choice":"string","probabilities":{},"confidence":123}},"usage":{"input_tokens":123,"output_tokens":123}}

Ask Noul

Submit a yes/no probability question to Typesafe AI and receive the result under answers.result. Requires state, model, and questions in the request body.

Endpoint

URL: v1/systemone

Method: POST

Input

Argument Name	Type	Required	Description
state	string	Optional	Content to evaluate. Plain text, or a JSON string such as a serialized ticket, chat log, or record.
model	string	Optional	Model id or alias. Default jev-latest.
questions	object	Optional	Parameter for Ask Noul
questions.result	object	Required	Result of the operation
questions.result.type	string	Required	Type of the resource
questions.result.instructions	string	Required	Yes/no question to evaluate. The answer is P(yes) from 0 to 1.
questions.result.criteria	object	Optional	Optional. Bind an object with keys "true" (what a yes near 1 means) and "false" (what a no near 0 means).

Input Example

{"state":"string","model":"jev-latest","questions":{"result":{"type":"noul","instructions":"string","criteria":{}}}}

Output

Parameter	Type	Description
status_code	integer	HTTP status code of the response
reason	string	Response reason phrase
model	string	Output field: model
answers	object	Output field: answers
answers.result	object	Result of the operation
answers.result.type	string	Type of the resource
answers.result.noul	number	Probability the answer is yes, from 0 to 1.
usage	object	Output field: usage
usage.input_tokens	integer	Input data for the action
usage.output_tokens	integer	Output field: usage.output_tokens

Output Example

{"model":"string","answers":{"result":{"type":"string","noul":123}},"usage":{"input_tokens":123,"output_tokens":123}}

Ask Score

Submit a score question to Typesafe AI and receive the result under answers.result by providing state, model, and questions in the request body.

Endpoint

URL: v1/systemone

Method: POST

Input

Argument Name	Type	Required	Description
state	string	Optional	Content to evaluate. Plain text, or a JSON string such as a serialized ticket, chat log, or record.
model	string	Optional	Model id or alias. Default jev-latest.
questions	object	Optional	Parameter for Ask Score
questions.result	object	Required	Result of the operation
questions.result.type	string	Required	Type of the resource
questions.result.instructions	string	Required	What the model should rate along the ordered criteria levels.
questions.result.criteria	array	Required	Ordered level descriptions, from the low end to the high end. At least 2 levels and at most 10.

Input Example

{"state":"string","model":"jev-latest","questions":{"result":{"type":"score","instructions":"string","criteria":["string"]}}}

Output

Parameter	Type	Description
status_code	integer	HTTP status code of the response
reason	string	Response reason phrase
model	string	Output field: model
answers	object	Output field: answers
answers.result	object	Result of the operation
answers.result.type	string	Type of the resource
answers.result.score	number	Probability-weighted value across the levels. Can fall between levels.
answers.result.legend	object	Level index, as a string key, mapped to its description.
answers.result.probabilities	object	Level index, as a string key, mapped to its probability. Values sum to 1.
answers.result.confidence	number	How certain the model is, from 0 to 1.
usage	object	Output field: usage
usage.input_tokens	integer	Input data for the action
usage.output_tokens	integer	Output field: usage.output_tokens

Output Example

{"model":"string","answers":{"result":{"type":"string","score":123,"legend":{},"probabilities":{},"confidence":123}},"usage":{"input_tokens":123,"output_tokens":123}}

Evaluate

Evaluate a given state against a set of typed questions using Typesafe AI and return answers mapped to each question.

Endpoint

URL: v1/systemone

Method: POST

Input

Argument Name	Type	Required	Description
state	string	Optional	Content to evaluate. Plain text, or a JSON string such as a serialized ticket, chat log, or record.
model	string	Optional	Model id or alias. jev-latest is the current stable Jev release. See GET /v1/models.
questions	object	Optional	Bind an object expression. Each value is a noul, choice, or score question with type and instructions. Noul criteria uses "true" and "false". Choice criteria is an option map of 2 to 255 entries. Score criteria is an ordered array of 2 to 10 level descriptions.

Input Example

{"state":"string","model":"jev-latest","questions":{}}

Output

Parameter	Type	Description
status_code	integer	HTTP status code of the response
reason	string	Response reason phrase
model	string	Output field: model
answers	object	One answer per question id. Noul has noul (0 to 1). Choice has choice, probabilities, and confidence. Score has score, legend, probabilities, and confidence.
usage	object	Output field: usage
usage.input_tokens	integer	Input data for the action
usage.output_tokens	integer	Output field: usage.output_tokens

Output Example

{"model":"string","answers":{},"usage":{"input_tokens":123,"output_tokens":123}}

List Models

Retrieve a list of model aliases and models accessible to your Typesafe AI account.

Endpoint

URL: v1/models

Method: GET

Output

Parameter	Type	Description
status_code	integer	HTTP status code of the response
reason	string	Response reason phrase
models	array	One entry per model or alias this account may send in the model field.
models.name	string	Model id or alias accepted by the model field.
models.description	string	What the model is for.
models.release_date	string	When the model or alias was released.

Output Example

{"models":[{"name":"Example Name","description":"string","release_date":"string"}]}

Response Headers

Header	Description	Example
Content-Type	The media type of the resource	application/json
Date	The date and time at which the message was originated	Thu, 01 Jan 2024 00:00:00 GMT