IBM Bob goes self-hosted: an AI coding agent behind your firewall
IBM Bob, IBM's AI coding agent, now runs self-hosted, on-prem, and air-gapped with Nemotron or Poolside models. What it means for where your source code goes.
Most AI coding tools work the same way: your source code leaves your network, a vendor's model reads it, and an answer comes back. IBM Bob self-hosted, announced this week, breaks that pattern. SiliconANGLE reports that IBM's agentic development platform now runs on-premises, in private and sovereign clouds, and in air-gapped environments. IBM is aiming at banks and governments. The question it raises applies to a ten-person shop too: where does your code go when an agent reads it?
What actually happened
Bob is IBM's platform for applying AI agents to the whole delivery cycle: understanding an existing app, planning changes, writing code, and checking results. Until now, it was a hosted service. Per CIO Dive, the new option lets companies run it next to code and data they keep on their own infrastructure.
The detail that matters is the model layer. SiliconANGLE says self-hosted Bob runs with Nvidia Nemotron and Poolside Laguna models. It can also call external model services in a hybrid setup, and customers can bring licenses for models they already pay for. IBM says it is available now. IBM did not publish pricing or hardware requirements.
IBM's pitch leans on its own research: a June 2026 IBM Institute for Business Value report found 68% of executives surveyed call data residency and sovereignty requirements a challenge. CIO Dive quotes Neel Sundaresan, IBM's general manager of AI and automation, saying organizations need AI that operates inside environments they control.
Why it matters for your business
You probably will not run Bob. You should still ask the question. Every AI coding tool your team uses sends something somewhere. Know which tools read your full repo, where that code is processed, and what the vendor keeps. If you hold client code under an NDA, the answer matters for your contracts.
The model is becoming a swappable part. Bob shipping with two model families plus bring-your-own is the useful signal. A coding agent that only works with one vendor's model locks you to that vendor's price and uptime. Prefer tools where the harness and the model are separate choices.
Self-hosted means self-patched. Running an AI agent inside your network moves the security job to you. Somebody owns updates, access control, and logs. Budget for that before you call self-hosting cheaper.
Key takeaways
- IBM Bob now runs on-prem, in private or sovereign clouds, and air-gapped
- Supported models include Nvidia Nemotron and Poolside Laguna, plus bring-your-own licenses
- IBM disclosed no pricing or hardware requirements
- Audit which AI coding tools read your repos and where that code is processed
- Pick agent tools where the model is a swappable part, and plan who patches self-hosted tools
Want AI dev tooling that keeps your code where you choose? We set up coding agents and automation with a model you can swap and data paths you can explain to a client. See what we build, or tell us about your setup.
Sources: SiliconANGLE, CIO Dive.
- #ibm-bob
- #ai-coding-agents
- #self-hosted-ai
- #data-sovereignty
- #developer-tools
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
Zammad CVE-2026-102489 on CISA KEV: patch your helpdesk
CISA added two exploited Zammad flaws to KEV. Chained, they turn a hijacked session into root on your helpdesk server. Upgrade to Zammad 7.2.0 and check for compromise.
Read itTA419 faked an Anthropic exec to phish: MFA codes didn't help
Proofpoint says China-aligned TA419 impersonated an Anthropic employee and used browser-in-browser phishing to steal MFA codes and sessions. Move to passkeys.
Read it