Allow administrators/users to mandate a specific model

IBM Bob intelligently selects AI models depending on the task, which is a useful default behaviour. However, enterprise customers should also have the ability to require a specific model for certain tasks, projects, repositories, or categories of operations.

For example, an organization could define that:

• a specific model must be used for code generation;
• another approved model must be used for code analysis;
• sensitive projects may only use models approved by the organization's AI governance team;
• some tasks must use a locally hosted model.

This capability will become particularly important for customers interested in an on-premises deployment of Bob, where organizations may want strict control over which models can process their source code and development data.

The model selection policy could potentially be configured at different levels, for example:

organization → project → repository → task.

Bob could continue to automatically select the best model by default, while administrators could override this behaviour where governance or technical requirements demand it.

Business value

Model selection policies would:

• support enterprise AI governance requirements;
• allow organizations to use only approved models;
• provide better control over data residency and sensitive source code;
• facilitate adoption in regulated environments;
• allow customers to balance quality, performance, cost, and sovereignty;
• be particularly valuable for future on-premises deployments where model control is a key architectural consideration.

Automatic model routing should remain Bob's default strength, but enterprise customers also need the ability to say: for this workload, use this model.