IBM Bob goes air-gapped: IBM's agentic coding platform now runs inside the data center

IBM announced on Thursday self-hosted deployment for IBM Bob, its agentic software development platform, allowing organizations to run AI-assisted software development on-premises, in private clouds, sovereign clouds, and air-gapped environments. The move targets the parts of the economy where sensitive source code, regulated data, and mission-critical infrastructure make public AI platforms a non-starter.

Bob is IBM's bet that AI coding goes well beyond code completion. The company positions it as an agentic platform for software delivery and modernization — the migrations, refactoring, and maintenance work that quietly consumes enterprise engineering budgets.

What “self-hosted Bob” actually means#

The new option brings Bob to customer-managed infrastructure. Enterprises can run supported models on premises — including in air-gapped environments — using models they have licensed themselves, or connect Bob to external model services through hybrid configurations. The design point is explicit: unlike most AI coding tools, which require moving data to external services, Bob runs where the code and data already live.

Sovereignty is the product#

“The future of enterprise AI will depend on security, governance and sovereignty,” said Neel Sundaresan, IBM's GM of AI and Automation. “Organizations need AI that operates inside environments they have control over, especially when working with sensitive code and regulated data. Bob's self-hosted deployment provides a way for enterprises to bring agentic AI directly to those environments so they can benefit from the technology while maintaining security, compliance, and operational control.”

The numbers behind the framing: Futurum Research projects hybrid and edge deployments will capture 44% of the AI infrastructure market by 2030 as organizations seek sovereign control alongside ecosystem connectivity, with public cloud share declining to 46% in the same period. And IBM's own Institute for Business Value reported in June that 68% of surveyed executives find meeting data residency and sovereignty requirements across geographies challenging.

Why air-gapped AI matters now#

The context is simple: agentic AI is most valuable where it has deep access — reading whole repositories, touching build pipelines, migrating legacy code. That access is precisely what regulated firms cannot grant to an external API. Air-gapped deployment removes the dilemma: the agent gets full repository context because nothing leaves the building.

It also fits the enterprise estate as it actually exists. Modernization projects rarely live in a single cloud; applications and data are spread across on-premises systems, private clouds, and other controlled environments. Most AI coding tools were designed for cloud-native setups, which leaves a large share of enterprise software work untouched by agentic AI.

The competitive read: IBM is betting the enterprise AI coding race will not be won on model quality alone but on deployment flexibility. Cloud-native agents are excellent when code can go to the cloud. For banks, telcos, healthcare, and governments, the answer has been “no AI agents.” Self-hosted Bob turns that answer into “yes — ours.”

What to watch#

Three things decide whether this lands. First, which models enterprises actually run with Bob in air-gapped setups — licensed on-premises models must be good enough for agentic work, a higher bar than chat. Second, whether regulated-industry pilots convert into production deployments, where compliance audits — not demos — decide. Third, whether cloud-native competitors answer: self-hosting is a gap in most AI coding stacks, and IBM just named it the feature.