IBM brings Bob to self-hosted environments for sensitive software projects

IBM has announced a self-hosted deployment option for Bob, its agentic software development platform. The October 1 announcement says organizations can deploy it on premises, in private or sovereign clouds, and in air-gapped environments.

The change addresses a concrete obstacle for development teams handling proprietary code or regulated information: where their software context goes when they use an AI tool. IBM's announcement says the self-hosted option brings development and modernization capabilities into environments where that code and data already reside. These are IBM's product claims, rather than results from a Franklin deployment test.

Customer-managed infrastructure changes the deployment choice

IBM describes Bob as a platform for software delivery and modernization, extending beyond code generation. The release focuses on giving enterprises control over data residency, security policies and development workflows while using those capabilities.

That focus suits organizations whose applications span existing on-premises systems and private clouds. Moving a development task into a public AI service may require permission to transfer the surrounding application context, not only the line of code being edited. IBM presents self-hosting as a way to bring AI into the controlled environment instead.

Deployment location is one part of that decision. An organization still needs to establish which information the system can access and which actions it may perform. IBM's description of customer control should not be mistaken for independent verification that a particular installation satisfies its owner's regulatory or security obligations.

Local models and hybrid configurations remain distinct

The announcement describes two model arrangements. Customers can run supported models they have licensed on premises, including in an air-gapped environment. They can also connect Bob to supported external model services through a hybrid configuration.

Those arrangements carry different data-flow implications. A self-hosted application that calls an external model service is not the same deployment as one whose model runs within an isolated environment. Teams assessing Bob should specify both the application's location and the model's location, rather than treating self-hosted as a complete answer about where processing occurs.

IBM limits the announcement to supported models. It does not say customers can substitute any model they choose. The captured release also does not enumerate the supported model versions, hardware requirements, deployment price or measured performance of an air-gapped installation. Procurement and engineering teams would need those details before committing to an environment.

IBM frames the release around residency requirements

IBM cites its Institute for Business Value report, published in June 2026, in which 68% of surveyed executives said meeting residency and sovereignty requirements across geographies was challenging. The figure describes the surveyed executives' experience; it does not measure how many organizations Bob will make compliant.

The release also cites a Futurum Research projection that hybrid and edge deployments will account for 44% of the AI infrastructure market by 2030, with public cloud at 46%. That is a forecast referenced by IBM, not a current market-share measurement or a guarantee of demand for this product.

Neel Sundaresan, IBM's general manager of AI and Automation, frames the release around operating AI inside environments customers control. The operational question for a buyer is narrower: can the supported deployment handle the required software work while respecting the organization's specific access and data-handling rules?

Bob's new option gives those buyers a deployment path to investigate. The announcement establishes the offered environments and model arrangements, but leaves the outcome of an individual modernization project to be demonstrated through its own evaluation.