BastionGPT (Preview)

BastionGPT is a HIPAA-compliant AI assistant built for healthcare by Bastion Intelligence. Generate text with AI chat completions, ask questions about uploaded documents, and transcribe audio recordings with speaker separation. Customer content is never used to train AI models. A BastionGPT API subscription key is required.

This connector is available in the following products and regions:

Service	Class	Regions
Copilot Studio	Premium	All Power Automate regions except the following: - US Government (GCC) - US Government (GCC High) - China Cloud operated by 21Vianet - US Department of Defense (DoD)
Logic Apps	Standard	All Logic Apps regions except the following: - Azure Government regions - Azure China regions - US Department of Defense (DoD)
Power Apps	Premium	All Power Apps regions except the following: - US Government (GCC) - US Government (GCC High) - China Cloud operated by 21Vianet - US Department of Defense (DoD)
Power Automate	Premium	All Power Automate regions except the following: - US Government (GCC) - US Government (GCC High) - China Cloud operated by 21Vianet - US Department of Defense (DoD)

Contact
Name	Bastion Intelligence Support
URL	https://support.bastiongpt.com
Email	support@bastiongpt.com

Connector Metadata
Publisher	Bastion Intelligence
Website	https://bastiongpt.com
Privacy policy	https://bastionintelligence.com/privacy
Categories	AI;Productivity

BastionGPT

Publisher: Bastion Intelligence

BastionGPT is a HIPAA-compliant AI assistant built for healthcare by Bastion Intelligence (FortaTech Security, LLC). This connector lets you generate text with AI chat completions, ask questions about uploaded documents, and transcribe audio recordings with speaker separation, all from your automations and agents. Customer content is never used to train AI models.

Prerequisites

A BastionGPT API subscription. Request one at https://bastiongpt.com/api-signup. Keys are typically issued the same business day.
Your BastionGPT API subscription key. Keep it confidential and rotate it if you suspect it has been exposed.

Obtaining Credentials

When you create a connection, paste your BastionGPT API subscription key into the API Key field. The connector sends the key in the key request header on every call. No other configuration is required.

Supported Operations

Ask a question

The simplest way to use BastionGPT. Sends a single question and returns the answer as plain text.

Question (required): what you want to ask or have generated.
Instructions (optional): guidance that shapes the response, such as the role to adopt, the audience, or the output format. When omitted, the default BastionGPT healthcare assistant persona is used.
Document ID (optional): the ID returned by Upload a document, to ask about that document.
Max Tokens (required, default 1000) and Temperature (advanced, default 0).

The response is flat: Answer, Finish Reason, Response ID, and Prompt/Completion/Total Tokens. Drag Answer directly into the next step.

Create a chat completion

Sends a multi-turn conversation to BastionGPT and returns an AI-generated text response. Use this when you need to send earlier turns or full control over the messages array; for a single question, Ask a question is simpler.

Messages (required): the conversation, oldest first. Each message has a Role (system, user, or assistant) and Content. Include earlier user and assistant turns to keep context. An optional system message as the first item overrides the default BastionGPT healthcare assistant persona.
Max Tokens (required, default 1000): the maximum number of tokens to generate.
Temperature (advanced, default 0): 0.0 to 1.0. Lower values give more consistent output.
Document ID (optional): the ID returned by Upload a document, to ask questions about that document.

The response contains Choices (an array with one item whose Message Content holds the generated text), a Response ID, a Created timestamp, and Usage token counts. To read the generated text in an expression, use:

first(outputs('Create_a_chat_completion')?['body/choices'])?['message']?['content']

Upload a document

Uploads a single document (PDF, plain text, HTML, CSV, Markdown, PNG, or JPEG; 10 MB maximum) and returns a Document ID. Supply the File content (for example the File Content output of a SharePoint or OneDrive Get file content action) and the File name (for example File name with extension). Documents are retained for 30 days after upload and can be referenced by any number of chat completion requests during that time. A document ID is owned by the API subscription key that uploaded it.

Submit audio for transcription

Uploads an audio recording (WAV, MP3, M4A, OGG, WEBM, and other common formats) for asynchronous transcription with speaker separation. Supply the File content and the File name. Returns a Transcript ID immediately, plus the audio File Duration in seconds.

Get a transcript

Retrieves a transcript by Transcript ID. The operation always succeeds with a Status field:

processing: the transcript is not ready yet. Segments is empty and Text is blank. Wait 10 to 20 seconds and call the operation again. A typical pattern is a Do until loop that repeats until Status equals completed, with a Delay of 15 seconds inside the loop.
completed: Segments holds the speaker-separated transcript (each with Timestamp in HH:MM:SS, Speaker, and Text), and Text holds the full transcript with one line per segment, ready to drop into an email, a document, or a record.

If the transcription failed or the transcript ID is missing, the operation returns an error with the service's message.

Getting Started

Request an API subscription at https://bastiongpt.com/api-signup and receive your key.
Add a BastionGPT action to your automation or agent and create a connection with your API key.
For a first test, use Ask a question with the Question Summarize the purpose of a SOAP note in two sentences. and Max Tokens 200, then use Answer in a following step.
To work with a file, use Upload a document first, then pass the returned Document ID to Ask a question or Create a chat completion.
To transcribe audio, use Submit audio for transcription, then poll Get a transcript until Status is completed.

Known Issues and Limitations

Max Tokens is required. Requests without it are rejected by the service.
One document per chat completion request. Documents expire 30 days after upload; upload again to get a new ID.
Chat completion size limits. Request body 10 MB; about 1.6 million characters per message and about 2.5 million characters of total message content. Very large prompts are automatically served by a large-context model.
Get a transcript reports processing for unknown transcript IDs as well as for in-progress transcriptions. If the status stays processing well beyond the audio duration, verify the transcript ID. Give Do until loops a sensible limit.
Streaming responses are not exposed by this connector. Use the non-streaming chat completion operation.
Rate limits depend on your subscription tier. On status code 429, wait and retry with exponential backoff.
File inputs need a content type. File content from SharePoint, OneDrive, Outlook attachments, or forms carries its content type automatically. If you build file content yourself from base64, use dataUriToBinary(concat('data:<content type>;base64,', <base64>)) so the service can recognize the file type.
Protected health information. BastionGPT is designed to process PHI under a Business Associate Agreement with Bastion Intelligence. Review your organization's policies before sending PHI through automations, and consider enabling secure inputs and outputs on actions so PHI is not written to run history.

Frequently Asked Questions

Which AI model answers my request?

BastionGPT automatically selects an appropriate enterprise-grade AI model for each request. The response format is the same regardless of the model used.

Is my data used to train AI models?

No. Customer content is never used to train AI models.

Where can I get help?

Visit https://support.bastiongpt.com or contact support@bastiongpt.com.

Creating a connection

The connector supports the following authentication types:


Default	Parameters for creating connection.	All regions	Not shareable

Default

Applicable: All regions

Parameters for creating connection.

This is not shareable connection. If the power app is shared with another user, another user will be prompted to create new connection explicitly.

Name	Type	Description	Required
API Key	securestring	Your API subscription key. Request one at bastiongpt.com/api-signup	True

Actions

Ask a question	Sends a single question to BastionGPT and returns the AI-generated answer as plain text. Optionally add instructions that shape the response, or supply a document ID to ask about a previously uploaded document. Use this operation for most scenarios; use the chat completion operation when you need to send a multi-turn conversation.
Create a chat completion	Sends a multi-turn conversation to BastionGPT and returns an AI-generated text response. Include earlier user and assistant turns in the messages to keep context, add a system message as the first item to set instructions, or supply a document ID to ask questions about a previously uploaded document. For a single question, the ask a question operation is simpler.
Get a transcript	Retrieves a transcript by its transcript ID. While transcription is still in progress, the status is processing and the segments are empty; wait 10 to 20 seconds and call the operation again. When the status is completed, the segments contain the speaker-separated transcript and the text contains the full transcript.
Submit audio for transcription	Uploads an audio recording for asynchronous transcription with speaker separation and returns a transcript ID immediately. Use the transcript ID with the get transcript operation to retrieve the completed transcript. Common audio formats such as WAV, MP3, M4A, OGG, and WEBM are accepted.
Upload a document	Uploads a single document and returns a document ID that can be supplied to the chat completion operation. Documents are retained for 30 days after upload and can be referenced by any number of chat completion requests during that time. Supported types are PDF, plain text, HTML, CSV, Markdown, PNG, and JPEG, up to 10 MB.

Ask a question

Operation ID:: AskQuestion

Sends a single question to BastionGPT and returns the AI-generated answer as plain text. Optionally add instructions that shape the response, or supply a document ID to ask about a previously uploaded document. Use this operation for most scenarios; use the chat completion operation when you need to send a multi-turn conversation.

Parameters

Name	Key	Required	Type	Description
Question	question	True	string	The question or request to send to BastionGPT.
Instructions	instructions		string	Optional instructions that shape the response, such as the role to adopt, the audience, or the output format. When omitted, the default BastionGPT healthcare assistant persona is used.
Document ID	document_id		string	The optional ID of a document previously uploaded with the upload a document operation. The document is attached to the request so the answer can use its content.
Max Tokens	max_tokens	True	integer	The maximum number of tokens to generate in the answer. Must be 1 or greater.
Temperature	temperature		float	The sampling temperature from 0.0 to 1.0. Lower values produce more consistent output and higher values produce more varied output.

Returns

Body: AskQuestionResponse

Create a chat completion

Operation ID:: ChatCompletion

Sends a multi-turn conversation to BastionGPT and returns an AI-generated text response. Include earlier user and assistant turns in the messages to keep context, add a system message as the first item to set instructions, or supply a document ID to ask questions about a previously uploaded document. For a single question, the ask a question operation is simpler.

Parameters

Name	Key	Required	Type	Description
Role	role	True	string	The author of the message. Use system for instructions, user for the person's input, and assistant for earlier AI responses.
Content	content	True	string	The text content of the message.
Max Tokens	max_tokens	True	integer	The maximum number of tokens to generate in the response. Must be 1 or greater.
Temperature	temperature		float	The sampling temperature from 0.0 to 1.0. Lower values produce more consistent output and higher values produce more varied output.
Document ID	document_id		string	The optional ID of a document previously uploaded with the upload a document operation. The document is attached to the request so the response can use its content. One document per request.

Returns

Body: ChatCompletionResponse

Get a transcript

Operation ID:: GetTranscript

Retrieves a transcript by its transcript ID. While transcription is still in progress, the status is processing and the segments are empty; wait 10 to 20 seconds and call the operation again. When the status is completed, the segments contain the speaker-separated transcript and the text contains the full transcript.

Parameters

Name	Key	Required	Type	Description
Transcript ID	transcriptId	True	string	The transcript ID returned by the submit audio for transcription operation.

Returns

Body: TranscriptResponse

Submit audio for transcription

Operation ID:: Transcribe

Uploads an audio recording for asynchronous transcription with speaker separation and returns a transcript ID immediately. Use the transcript ID with the get transcript operation to retrieve the completed transcript. Common audio formats such as WAV, MP3, M4A, OGG, and WEBM are accepted.

Parameters

Name	Key	Required	Type	Description
Audio File	file	True	file	The audio recording to transcribe. Common formats such as WAV, MP3, M4A, OGG, and WEBM are accepted.

Returns

Body: TranscribeResponse

Upload a document

Operation ID:: UploadDocument

Uploads a single document and returns a document ID that can be supplied to the chat completion operation. Documents are retained for 30 days after upload and can be referenced by any number of chat completion requests during that time. Supported types are PDF, plain text, HTML, CSV, Markdown, PNG, and JPEG, up to 10 MB.

Parameters

Name	Key	Required	Type	Description
File	file	True	file	The document to upload. Supported types are PDF, plain text, HTML, CSV, Markdown, PNG, and JPEG, up to 10 MB.

Returns

Body: DocumentUploadResponse

Definitions

AskQuestionResponse

Name	Path	Type	Description
Answer	answer	string	The AI-generated answer text.
Finish Reason	finish_reason	string	The reason generation stopped. The value stop means a natural end, max_tokens means the token limit was reached and the answer may be truncated, and content_filter means content was withheld by safety filtering.
Response ID	response_id	string	The unique identifier of this response.
Prompt Tokens	prompt_tokens	integer	The number of tokens in the input, including any instructions.
Completion Tokens	completion_tokens	integer	The number of tokens generated in the answer.
Total Tokens	total_tokens	integer	The sum of prompt tokens and completion tokens.

TranscriptResponse

Name	Path	Type	Description
Status	status	string	The transcription status. The value processing means the transcript is not ready yet and completed means the segments and text are available.
Segments	segments	array of TranscriptSegment	The speaker-separated transcript segments in order. Empty while the status is processing.
Text	text	string	The full transcript as text, one line per segment with its timestamp and speaker. Empty while the status is processing.

TranscriptSegment

Name	Path	Type	Description
Timestamp	timestamp	string	The segment start time in HH:MM:SS format.
Speaker	speaker	string	The speaker label assigned by speaker separation.
Text	text	string	The transcribed text for this segment.

ChatCompletionResponse

Name	Path	Type	Description
Choices	choices	array of Choice	The generated responses. This array contains one item.
Response ID	id	string	The unique identifier of this response.
Created	created	integer	The Unix timestamp in seconds when the response was created.
Usage	usage	Usage	The token accounting for the request.

Choice

Name	Path	Type	Description
Finish Reason	finishReason	string	The reason generation stopped. The value stop means a natural end, max_tokens means the token limit was reached and the response may be truncated, and content_filter means content was withheld by safety filtering.
Content	message.content	string	The generated text.

Usage

The token accounting for the request.

Name	Path	Type	Description
Prompt Tokens	promptTokens	integer	The number of tokens in the input, including the system message.
Completion Tokens	completionTokens	integer	The number of tokens generated in the response.
Total Tokens	totalTokens	integer	The sum of prompt tokens and completion tokens.

DocumentUploadResponse

Name	Path	Type	Description
Document ID	document_id	string	The identifier of the uploaded document, valid for 30 days after upload. Supply it to the chat completion operation.

TranscribeResponse

Name	Path	Type	Description
Transcript ID	transcriptId	string	The identifier used to retrieve the completed transcript.
File Duration	fileDuration	string	The duration of the submitted audio in seconds, as a decimal string.