AI
| Trisotech Digital Enterprise Suite AI features rely on external AI providers that must be configured. These features should be considered incubating and may change or be removed without warning. Different AI providers and models will yield different results. |
The Trisotech Digital Enterprise Suite allows users to leverage existing AI models published by AI providers through the use of various model providers.
By default, interactions with AI models are not configured, and the administrator must configure and enable each one.
Currently, the following providers are supported.
Provider |
Description |
OpenAI |
The models available from OpenAI can be found in their documentation: https://developers.openai.com/api/docs/models. |
OpenAI Compatible |
The OpenAI API is popular, and many smaller providers offer compatible APIs. This option allows connections to those providers. The models available will depend on the provider used. An additional API URL is required. For example, the DeepSeek OpenAI-compatible API is https://api.deepseek.com/v1, and the Ollama-compatible API is http(s)://ollamahost:11434/v1. Refer to the provider documentation to obtain the OpenAI-compatible base URL. |
Google Gemini |
The models available from Google Gemini can be found in their documentation: https://ai.google.dev/gemini-api/docs/models. |
Anthropic |
The models available from Anthropic can be found in their documentation: https://platform.claude.com/docs/en/about-claude/models/overview. |
GitHub Models |
The models available from GitHub can be found in their documentation: https://github.com/marketplace?type=models. |
Most model providers require some type of identification, most often through an API key, which can be entered as an Identity in the Digital Enterprise Suite.
Each AI feature can be restricted to specific users through group membership. You can create a group containing the users who should have access to AI features and then select this group in the configuration.
AI Performers
AI Performers can be created and used as performers for user tasks in the Workflow Modeler.
It is possible to add ( New AI Performer ), edit ( ), or delete ( ) an AI Performer.
AI Performers can contain a system prompt to instruct the model on how to perform tasks. The model will also receive the description of the task to be performed and the task inputs. It will then generate the task outputs based on this information.
To perform their tasks, AI Performers can call tools. Tools are actions that can be performed by the model to obtain additional information. For example, a tool can call an external system to retrieve additional information. The AI model will determine when to call a tool and which parameters to use based on the information it receives about the task to perform. Two types of tools are currently available.
Tool Type |
Description |
MCP |
The Model Context Protocol (MCP) is a standard that allows systems to be exposed to AI models through a standardized API. Both HTTP Streaming and Server-Sent Events (SSE) are supported. When selecting an MCP endpoint, all tools exposed by that endpoint will be available for the AI Performer to use. It is also possible to specify a subset of the tools to use. Endpoints requiring security can also be used by defining an identity for the endpoint. |
Service |
Services deployed in the Service Library can be transparently exposed as tools for AI Performers. When selecting a service, you must specify the environment, group, artifact, and version of the service using the dropdown lists. |
| When using tools, the AI model will be free to call these services as needed. Make sure to consider the security implications of making these tools available to the performer and to users who will have access to this performer in the models they create. |
AI Model Generators
Two types of generators are currently available to help modelers create their models.
The model generator allows users to generate a model based on a description and iterate on the result with further instructions. The model is generated directly in the modeler canvas. Model generation performance, including both precision and generation time, depends on the AI model used, the prompt provided, and the complexity of the model being generated. It is recommended to start with simple models and prompts and iterate from there.
The model descriptor generates a textual description of a model based on its content. It is available at the model level for Workflow, Case, Decision, and Shared Data Models. The description is generated by analyzing the content of the model and providing a business-level description of that model. The description can help modelers understand the content of a complex model or serve as a basis for documentation.