Most AI coding assistants work by sending your code off to someone else’s cloud and trusting it comes back unharmed. For a bank or a government handling regulated or classified source, that is a non-starter. On 1 October, IBM made its pitch to exactly those customers: a version of its coding agent, IBM Bob, that runs entirely inside your own walls.
What changed
IBM introduced a self-hosted deployment for Bob, its agentic software-development platform, letting organisations run it on-premises, in private or sovereign clouds, and in fully air-gapped networks with no internet connection at all. The core features survive the move, including the BobShell interface and parallel tool calling, so this is not a stripped-down version traded for privacy. For models, IBM currently supports Nvidia’s Nemotron and Poolside’s Laguna running on the customer’s own hardware, with hybrid setups for teams that want to mix local and external models.
Who this is for
The target is blunt: banks, governments and other heavily regulated outfits that have sat out the AI-coding boom because their rules forbid sending proprietary code to a third-party AI cloud. “AI sovereignty” is the phrase IBM is leaning on, and for once it is doing real work rather than padding a slide. If your compliance team has been the reason you could not touch a cloud coding assistant, an agent that provably never leaves your network removes the objection.
Why it matters
This is the enterprise AI market maturing in real time. The first wave was about capability: who has the smartest model. The second wave, the one that unlocks the biggest and most conservative budgets, is about control: where the model runs, who can see the data, and whether you can switch it off. IBM is betting that a lot of serious money has been waiting on the sidelines for precisely this, and the market seemed to agree, with IBM shares up around 4 percent in pre-market trading on the news.
The catch
Running models on your own hardware is not free. You need the GPUs, the people to maintain them, and the appetite to own the whole stack rather than rent it. For most small companies, the convenience of a cloud agent still wins. But for the Fortune 500 bank or the defence ministry, owning the stack is the entire point, and the hardware cost is a rounding error against the risk of a leak.
What this means
If you work somewhere regulated and were told AI coding tools are off-limits, that answer may have changed this week. The broader signal for everyone else: expect “self-hosted” and “air-gapped” to become standard options across AI tooling, because the money that cares about control is bigger than the money chasing the last few benchmark points.
Sources: IBM (announcement, 1 Oct); MarkTechPost; AIwire; BigGo Finance.