An assistant and modules are created and modified via API call. This guide leads you through the calls you have to do and the settings and configurations that can be set.

Preferences

Token of a user / service-user which has an admin role assigned → <yourToken>
The base URL of your application → <baseUrl>!

To get the token and the URL check this: How to get a Token for our APIs

Assistant

Create a new assistant

This can be done with this cURL. Just replace the following placeholders:

<baseUrl> / <yourToken>
<assistantDefinition> → Definition of your assistant with corresponding modules (see example below)

curl --location --globoff 'https://gateway.<baseUrl>/chat/graphql' \
--header 'Content-Type: application/json' \
--header 'Authorization: Bearer <yourToken>' \
--data '{"query":"mutation CreateAssistant($input: AssistantCreateInput!) {\n    createAssistant(input: $input) {\n        id\n        name\n        fallbackModule\n        languageModel\n        chatUpload\n        settings\n        modules{\n            id\n            name\n            configuration\n            weight\n            isExternal\n            }\n    }\n}","variables":<assistantDefinition>}'

Update an assistant

To update some settings of an assistant, use the following cURL. Update these placeholders:

<baseUrl> / <yourToken>
<assistantId>
"param1": "<paramValue>" → contains variables that should be updated. Variables that are not contained in input remain unchanged

curl --location --globoff 'https://gateway.<baseUrl>/chat/graphql' \
--header 'Content-Type: application/json' \
--header 'Authorization: Bearer <yourToken>' \
--data '{"query":"mutation UpdateAssistant($updateAssistantId: String!, $input: AssistantUpdateInput!) {\n  updateAssistant(id: $updateAssistantId, input: $input) {\n    name\n    id\n    languageModel\n    settings\n    chatUpload\n    modules {\n      id\n      name\n      configuration\n    }\n  }\n}","variables":{"updateAssistantId":"<assistantId>","input":{"param1": "<paramValue>"}}}'

An example command to update the languageModelused to perform function calling and module selection can be updated with the following curl command:

curl --location 'https://gateway.<baseUrl>/chat/graphql' \
--header 'Content-Type: application/json' \
--header 'Authorization: Bearer <yourToken>' \
--data '{"query":"mutation UpdateAssistant($updateAssistantId: String!, $input: AssistantUpdateInput!) {\n      updateAssistant(id: $updateAssistantId, input: $input) {\n            name\n            id    \n            languageModel    \n            settings    \n            chatUpload    \n            modules {\n                id\n                name\n                configuration   \n            }\n       }\n}","variables":{"updateAssistantId":"<assistantID>","input":{"languageModel":"AZURE_GPT_4_0613"}}}'

Delete an assistant

To delete an assistant, use the following cURL. Update these placeholders:

<baseUrl> / <yourToken>
<assistantId>

Assistant variables and settings

The list contains various parameters that need to be set when an assistant is created/updated.

Variable	Description	Options	optional / required

Variable	Description	Options	optional / required
`name`	Name of assistant		required
`fallbackModule`	fallbackModule if module selection could not find a suitable module		required
`languageModel`	Language module used for module selection	see below section with GPT-models default: `AZURE_GPT_35_TURBO`	optional
`chatUpload`	Enable/disable upload of documents into chat	`DISABLED` (default) `ENABLED`	optional
`settings` Collection of various setting	`showPdfHighlighting`: Open references in same chat window on the right side. Referenced section is highlighted in the PDF	`showPdfHighlighting` `false` (default) `true`	optional
	`modelChoosing`: type of module selection	`modelChoosing` `BY_FUNCTION_CALL` `BY_PROMPT` (default)	optional
	`isPinned`: flag if a space is pinned (marked with a star). This is only relevant for spaces	`isPinned` `false` (default) `true`	optional
`modules`	List of modules to create. See Module dependent configurations for details about configuration of a module		required

Example input for an Internal Knowledge assistant containing two modules (SearchInVectorDB, Translate).

You can find more about assistants here: Assistants

Module

Create a module

This can be done with this cURL. Just replace the following placeholders:

<baseUrl> / <yourToken>
<assistantId> → ID of the assistant where the module should be created
<moduleDefinition> → Definition of the new module (see example below)

Update a module

This can be done with this cURL. Just replace the following placeholders:

<baseUrl> / <yourToken>
<moduleId>
<updatesModule> → New set of module parameters

Delete a module

This can be done with this cURL. Just replace the following placeholders:

<baseUrl> / <yourToken>
<moduleId>

Module variables and configuration

Below is an example of parameters for a module:

configuration → depends on module. See section Module dependent configurations for more details
isExternal → flag if module is external (= developed with SDK)
weight → defines the order of modules in module selection prompt
toolDefinition → Description of module used for function calling. Here, the same structure as openAI is used (therefore, this parameter is called toolDefinition and not moduleDefinition)

Module dependent configurations

Module	Example	Description	Parameter	Options

Module	Description	Parameter	Options
`SearchInVectorDB`	GPT model to be used	`languageModel`: string	see below section with GPT-models default: `AZURE_GPT_35_TURBO`: GPT-35-turbo (0301)
	Scopes that the module can access	`scopeIds`: object
	RAG approach to search for chunks	`searchType`: string	`VECTOR`: semantic search (similarity of input and chunk embeddings) `COMBINED`: hybrid search combining vector and fulltext search
	Describing if chunks of same document are appended as individual sources to GPT content or merged to one source	`chunkedSources`: boolean	`true`: each chunk is a separate source `false`: chunks from the same document are merged to one source
	Scope restriction to documents that are uploaded. If no documents are uploaded, then scopes in scopeIds are relevant.	`scopeToChatUpload`: boolean	`true`: Scope restriction on (if document uploaded) `false`: Scope restriction off
	Flag that allows to include previous chat conversation in GPT-calls only if the new user input is a follow-up question	`historyIncluded`: boolean	`true`: History always included `false`: History only included for follow-up questions
	Max tokens used by sources and previous conversation	`maxTokens`: integer	Default value depends on the used `languageModel` `AZURE_GPT_35_TURBO`: 3000 `AZURE_GPT_35_TURBO_0613`: 3000 `AZURE_GPT_35_TURBO_16K`: 14000 `AZURE_GPT_4_0613`: 7000 `AZURE_GPT_4_32K_0613`: 30000 `AZURE_GPT_4_VISION_PREVIEW`: 7000 `AZURE_GPT_4_TURBO_1106`: 126000 `AZURE_GPT_4_TURBO_2024_0409`: 126000 `AZURE_GPT_4o_2024_0513`: 126000 `AZURE_GPT_4o_2024_0806`: 126000 `AZURE_GPT_4o_MINI_2024_0718`: 126000
	Specifies the primary language used for full-text search. This should match the predominant language of the documents in the knowledge centre.	`ftsSearchLanguage`: string	Default: `english`
`ContextMemorySearch`	GPT model to be used	`languageModel`: string
`ContextMemorySearch`	Describing if chunks of same document are appended as individual sources to GPT content or merged to one source	`chunkedSources`: boolean	`true`: each chunk is a separate source `false`: chunks from the same document are merged to one source
`DocumentSummarizer`	GPT model to be used	`languageModel`: string
`EmailWriter`	GPT model to be used	`languageModel`: string
`ExternalKnowledge`	GPT model to be used	`languageModel`: string
	Temperature (chatGPT)	`temperature`: number	Range: 0-1 Default: 0.5
	System prompt	`systemPromptExternalKnowledge`: string	Default system prompt is (depending of `languageModel`):
	Maximum number of user-assistant interactions taken into account in the history.	`maxHistoryInteraction`: number	Default: 2
`InvestmentResearchDocuments`
`InvestmentResearchTable`
`LunchSearchV4`	GPT model to be used	`languageModel`: string
`QueryTable`	GPT model to be used	`languageModel`: string
	tbd	`tableconfig`:
	tbd	`searchExamples`:
	tbd	`showTableReference`: boolean
`TranscriptInteraction`	GPT model to be used	`languageModel`: string
	Scope that the module can access	`scopeId`: string
	Max tokens used by sources and previous conversation	`maxTokens`: integer
	Name of excel template file, that will be filled with extracted values. Need to uploaded to the same scopeId	`templateName`: string
`Translate`	GPT model to be used	`languageModel`: string
`WhatsappWriter`	GPT model to be used	`languageModel`: string

Available GPT models

The list below contains available GPT-models with the corresponding name that has to be used in the configurations for the assistants and modules:

Model	Key

Model	Key
GPT-35-turbo (0301)	AZURE_GPT_35_TURBO
GPT-35-turbo (0613)	AZURE_GPT_35_TURBO_0613
GPT-35-turbo-16K (0613)	AZURE_GPT_35_TURBO_16K
GPT-4 (0613)	AZURE_GPT_4_0613
GPT-4-32K (0613)	AZURE_GPT_4_32K_0613
GPT-4-turbo (0409)	AZURE_GPT_4_TURBO_2024_0409
GPT-4o (2024-0513)	AZURE_GPT_4o_2024_0513
GPT-4o (2024-0806)	AZURE_GPT_4o_2024_0806
GPT-4o-mini (2024-0718)	AZURE_GPT_4o_MINI_2024_0718

GPT versions available in Preview mode (not recommended to be used in productive applications):

Model	Key

Model	Key
GPT-4-turbo (1106)	AZURE_GPT_4_TURBO_1106

You can find more about modules here: Modules

Author	@Fabian Schläpfer

Public Documentation

Assistant & module settings

Preferences

Assistant

Create a new assistant

Update an assistant

Delete an assistant

Assistant variables and settings

Module

Create a module

Update a module

Delete a module

Module variables and configuration

Module dependent configurations

Available GPT models