TaskSultan can use its own orchestration and model adapter for Scribe and Builder. Depending on the deployment, the model can be hosted by TaskSultan, supplied by the customer, or run locally.
#Examples of model choices
Examples include OpenAI-compatible models, Anthropic models, Google Gemini, DeepSeek and local models served through Ollama or another compatible endpoint. The exact models available depend on the deployment and the customer’s credentials, region and policy.
#Credit boundary
Using a third-party model for requirements extraction or first-draft generation consumes that provider’s usage, not UiPath’s AI credit pool. UiPath runtime features such as native AI services may still consume UiPath entitlements.
#Grounding and review
The agent receives the approved specification and relevant evidence, not the entire customer environment. Answers and generated code should cite their source requirements, surface uncertainty and remain reviewable before execution.