I'm Nitin Kumar, Lead Project Coordinator at Debugged Pro. I help software and IT agencies in India and the USA ship client work without guessing what vendors actually released. IBM Bob just got a self-hosted path that many regulated clients have been waiting for.

On October 1, 2026, IBM's newsroom said enterprises can now bring IBM Bob to on-premises, private-cloud, sovereign-cloud, and air-gapped environments. IBM's product announcement also frames the self-hosted option as generally available. That matters if your clients refuse to send source code to a public AI coding SaaS.

This guide sticks to IBM's newsroom post, the IBM.com announcement, and IBM Bob's own self-hosted docs. No invented benchmarks. No fake quotes. Just what delivery teams need to brief clients and size a pilot.

IBM Bob self-hosted product interface shown in IBM press materials — Feature image: IBM Bob interface from IBM Newsroom press materials. Credit: IBM.

What IBM Bob is, in plain terms

IBM calls Bob an agentic software development platform. It is not only autocomplete. IBM says Bob helps teams understand, plan, execute, and validate software development and modernization work across the lifecycle.

For agencies, that means Bob sits closer to delivery work than a chat window. Think IDE workflows, shell workflows, planning steps, and modernization packages for long-lived systems.

Until now, many buyers liked the pitch and still blocked the tool. Their security and compliance teams wanted the model and the code to stay inside infrastructure they control. Self-hosted Bob is IBM's answer to that block.

Why IBM Bob self-hosted showed up now

Neel Sundaresan, GM of AI and Automation at IBM, put the stake in the ground clearly: "The future of enterprise AI will depend on security, governance and sovereignty." He added that organizations need AI that operates inside environments they control, especially with sensitive code and regulated data.

IBM also cited outside and internal research that matches what we hear in client calls:

Futurum Research projects hybrid and edge deployments will capture 44% of the AI infrastructure market by 2030, while public cloud share falls to 46% in that timeframe.
IBM's Institute for Business Value report "The Calculus of AI Sovereignty" (June 2026) found 68% of surveyed executives say meeting data residency and sovereignty requirements across geographies is challenging.

For India delivery centers serving USA banks, insurers, and healthcare clients, those numbers are not abstract. Data residency clauses already kill cloud-only AI coding tools in security review.

How IBM Bob self-hosted differs from SaaS

IBM Bob docs compare the two modes in a simple table. The developer-facing tools stay familiar. The operations model flips.

Area	SaaS	Self-hosted
Infrastructure	Hosted and managed by IBM	Managed by the client on their OpenShift cluster
Operations	IBM handles upgrades, scaling, availability	Client owns lifecycle operations
Security controls	IBM-managed	Client configures networking, storage, and identity
Data residency	IBM-managed regions	Full control inside the client environment
Developer experience	Bob IDE extensions and bob-shell	Same Bob client tools and workflows

That last row is the agency-friendly part. Developers keep the same Bob IDE extensions and bob-shell experience. Platform teams take on cluster work. Your proposal should price both sides, not only seats.

IBM Bob home page banner from official press distribution — Body image: IBM Bob banner distributed with the PR Newswire release. Credit: IBM / PR Newswire.

Architecture facts delivery teams should memorize

OpenShift, not a mystery stack

Self-hosted Bob runs on Red Hat OpenShift Container Platform. Docs say it deploys as a standard OpenShift workload and can share a cluster with other apps. A dedicated cluster is not required, but many regulated clients will still prefer one.

IBM documents two primary namespaces:

Operator namespace for the Bob Operator and required dependency operators.
Instance namespace for Bob application services, databases, authentication services, and supporting workloads.

Supported OCP versions in the current system requirements are 4.20, 4.21, and 4.22. Worker nodes that schedule Bob workloads must be amd64 (x86_64). Mixed-architecture clusters are only fine if you pin Bob to amd64 nodes yourselves. Bob does not apply those constraints automatically.

Sizing numbers you can put in a proposal

IBM publishes aggregate Bob tenant footprints. These exclude OpenShift platform overhead, so do not treat them as full cluster size.

Bob Core: about 22.1 vCPU, 35.1 GiB memory, ~30 GiB storage.
Bob Core + RAG: about 38.1 vCPU, 69.1 GiB memory, ~62 GiB storage.
Bob Core + Z Understand: about 30.1 vCPU, 74.1 GiB memory, ~2288 GiB storage (benchmarking still in progress).
Bob Core + RAG + Z Understand: about 46.1 vCPU, 108.1 GiB memory, ~2320 GiB storage (benchmarking still in progress).

For production Bob Core, IBM also shows a raw footprint around 28.1 vCPU and 41.1 GiB, plus a headroom-adjusted target near 36.5 vCPU and 53.4 GiB. A minimum reference topology in the docs uses 3 control plane, 3 infrastructure, and 3 worker nodes. Treat that as planning material, not a one-size mandate.

Important ops note: security event logging and monitoring for self-hosted Bob are managed at the OpenShift platform level. Bob does not provide that layer. Your runbooks must cover log retention, SIEM shipping, and audit needs.

Model choices: self-hosted, hybrid, and BYOL

At GA, IBM says customers source and provide access to a supported model. The announcement groups models by deployment method.

Self-hosted models named by IBM

NVIDIA Nemotron
Poolside Laguna

Hybrid or private SaaS models named by IBM

Claude Sonnet 5.0
Claude Opus 4.8
Gemini 3.7 Flash
OpenAI GPT 5.6 Sol

IBM also describes a bring-your-own-license (BYOL) path for eligible licenses on the supported model list. That helps clients who already paid for model capacity and do not want a second commercial fight.

IBM's example is practical: a bank could keep core banking work on on-premises model inference, then use an approved external model for a less restricted workload. Developers still see one Bob experience while platform teams map inference to policy.

IBM says it plans to expand the supported model portfolio and add multi-model routing later. Do not sell routing as a current GA checkbox unless your IBM account team confirms it for that client.

IBM Bob AI coding agent leadspace artwork from IBM.com — Body image: IBM Bob AI coding agent leadspace from IBM's announcement page. Credit: IBM.

Premium packages agencies should ask about

Core Bob covers the shared IDE and shell experience. IBM also lists optional Premium Packages that matter for modernization deals:

Java Modernization for dedicated modernization workflows.
IBM i for platform-specific connectivity, tools, and IBM i knowledge.
IBM Z for enterprise-scale understanding and modernization on mainframe estates.

These packages are subject to licensing and deployment requirements. For India and USA agencies that already modernize Java, IBM i, or Z systems, this is where Bob can move from coding help to a funded transformation workstream.

What this means for India and USA agency delivery

Security reviews get a clearer yes path

Cloud-only coding agents often die in the first security meeting. Self-hosted Bob gives you a concrete alternative: code, context, and build artifacts can stay inside the customer-managed environment when you use a supported self-hosted model setup.

That does not remove due diligence. It changes the conversation from "we cannot use AI on this repo" to "which OpenShift project, which model path, and which audit controls."

Delivery roles change

Expect three workstreams in any serious pilot:

Platform: OpenShift capacity, namespaces, identity, storage, backups, upgrades.
Model: Nemotron, Laguna, or an approved hybrid endpoint, plus BYOL paperwork.
Delivery: repo access patterns, prompt and skill standards, human review gates, modernization backlog selection.

If you only staff developers, the pilot stalls on day two. Pair a platform lead with a delivery lead from the start.

India offshore + USA onshore patterns

Many of our India teams already work inside client VPNs and private clouds for USA programs. Self-hosted Bob fits that pattern better than a consumer coding agent tied to a public SaaS region.

Use clear data maps. Show which prompts, repo mirrors, and telemetry leave the client boundary. If hybrid models are in play, put that path in writing before coding starts.

For broader AI governance context that clients ask about in the same meetings, see our White House AI accord agency guide. For identity mistakes that still show up after tools are approved, see our AI agent credentials agency guide.

A practical 30-day pilot plan for agencies

Days 1-5: Confirm OCP version, amd64 capacity, and whether the client wants a shared or dedicated cluster. Pull IBM's system requirements into the architecture note.
Days 6-10: Pick one model path. Prefer fully self-hosted inference for the highest sensitivity repos. Document hybrid endpoints only where policy allows.
Days 11-15: Install Bob Core in a non-production project. Wire identity. Turn on OpenShift-level audit logging before inviting developers.
Days 16-22: Run a narrow backlog: one service, one modernization spike, or one high-churn module. Measure review time, defect escapes, and secrets handling.
Days 23-30: Decide expand, hold, or stop. If expanding, price operator ownership, package add-ons, and support hours separately from developer seats.

Keep the pilot boring on purpose. The win is a path security accepts, not a flashy demo that cannot leave the sandbox.

Client briefing script you can reuse

When a USA or India client asks why this release matters, keep the brief short:

IBM Bob is IBM's agentic software development platform for coding and modernization work.
Self-hosted deployment is now positioned for customer-managed OpenShift, including air-gapped setups.
Developers keep familiar Bob IDE and bob-shell workflows while platform teams control residency.
Supported model paths include on-prem models and approved hybrid private SaaS connections.
Optional Premium Packages target Java, IBM i, and IBM Z modernization once core is stable.

Read IBM's primary materials before you sell the story: the IBM Newsroom release, the IBM.com self-hosted announcement, and the Bob self-hosted overview docs.

Common mistakes to avoid

Selling SaaS pricing for a self-hosted job. Cluster ops, backups, and upgrades are real delivery cost.
Ignoring amd64 and OCP version gates. Unsupported clusters create week-long install thrash.
Skipping SIEM ownership. Bob expects you to handle security logs at the OpenShift layer.
Starting on the crown-jewel monolith. Prove the control plane on a medium-risk service first.
Forgetting package licensing. Java, i, and Z add-ons are not automatic with Core.

If your client also asks how agent runtimes should be sandboxed while you evaluate tools, our NVIDIA agent safety agency guide is a useful companion brief. For consulting help on delivery design, visit our software development consulting page or the main Debugged Pro blog.

Key Takeaways

IBM Bob self-hosted is IBM's October 1, 2026 answer for agentic development inside controlled infrastructure.
It targets on-premises, private-cloud, sovereign-cloud, and air-gapped OpenShift environments.
Developer tools stay consistent with SaaS Bob; operations and residency move to the client.
Plan capacity with IBM's published Bob Core and add-on footprints, plus OpenShift overhead.
Pick model paths early: self-hosted Nemotron or Laguna, or approved hybrid private SaaS models.
India and USA agencies should staff platform, model, and delivery leads for any pilot.

Where Debugged Pro fits

Debugged Pro works with agencies that need sober vendor briefs and delivery plans, not hype decks. If you are scoping an IBM Bob self-hosted pilot for a regulated India or USA client, we can help with architecture notes, backlog selection, and rollout checkpoints.

Start at contact us or browse more delivery guides on our blog. You can also review related services on our services page.

About the author: Nitin Kumar is Lead Project Coordinator at Debugged Pro. He coordinates delivery across India and USA agency programs, with a focus on practical AI tooling, secure software delivery, and client-ready documentation.