IBM Makes Bob Self-Hosted Option Generally Available for Air-Gapped Agentic Development

Bob covers the full software lifecycle: understanding code, planning work, executing changes and validating results. IBM said the self-hosted GA includes Bob running in customer-managed enterprise environments, the core Bob capabilities of the IDE experience, BobShell, parallel tool calling, the agent harness, skills and modes, self-hosted, air-gapped and hybrid model configurations, bring-your-own-license terms for eligible models on IBM's supported list, integration with supported IDE and infrastructure environments, and optional Premium Packages for Java Modernization, IBM i and IBM Z. Customers must source, license and host a supported model themselves. IBM has not published pricing and routes buyers to a demo request.

The model layer determines where inference runs. Bob self-hosted does not bundle a model; customers bring one from IBM's supported list. IBM groups that list by deployment method. In the self-hosted method, the customer hosts the model on its own infrastructure, and the supported models at GA are NVIDIA Nemotron and Poolside Laguna. In a hybrid or private SaaS configuration, the model is an external service reached through an approved private connectivity pattern, with supported models at GA including Claude Sonnet 5.0, Claude Opus 4.8, Gemini 3.7 Flash and OpenAI GPT 5.6 Sol. With a self-hosted model, code, development context and build artifacts can remain inside the customer environment. IBM gave the example of a bank running on-premises inference for core banking applications while routing a less restricted workload to an approved external model. Developers see a consistent core Bob experience in both cases, while platform and security teams control the architecture underneath. IBM also plans to expand the model portfolio and add multi-model routing, but MarkTechPost reported that this is roadmap rather than a shipped feature.

MarkTechPost compared Bob self-hosted with GitLab Duo Agent Platform Self-Hosted, Mistral Vibe for Code Enterprise and GitHub Copilot CLI BYOK. IBM, GitLab and Mistral offer on-premises or private cloud deployment, while GitHub Copilot CLI is CLI only and core Copilot runs on GitHub.com or GHE.com. For air-gapped use, IBM supports it with supported self-hosted models, GitLab with an offline license, Mistral with open-weight Devstral models, and GitHub with offline mode using a local model. Hybrid local plus external models are supported by IBM per workload, by GitLab through a hybrid AI Gateway, by Mistral through on-premises, private cloud or Mistral Cloud, and by GitHub through BYOK providers. When self-hosted, IBM lists NVIDIA Nemotron and Poolside Laguna; GitLab supports a list on vLLM, Bedrock or Azure OpenAI; Mistral offers its models with fine-tuning via Forge; GitHub allows any provider with tool calling and streaming. Developer surfaces are the IDE and BobShell for IBM, the GitLab UI, IDE and beta CLI for GitLab, CLI plus VS Code, JetBrains and Zed for Mistral, and the terminal for GitHub. Commercially, IBM sells through IBM sales as an enterprise offering, GitLab requires Premium or Ultimate plus an add-on, Mistral asks enterprises to contact sales, and GitHub says no Copilot license is needed with BYOK. MarkTechPost cited IBM, GitLab docs, Mistral Vibe, Mistral's coding stack, GitHub's changelog and GitHub Docs for the comparison.

The scope of the offerings differs. GitLab and GitHub bring agents to their own DevOps platforms. Mistral pairs its agent with its own open-weight models. IBM targets modernization of long-lived estates, especially Java, IBM i and mainframe code, inside environments that cannot touch the public internet. The self-hosted option is GA for on-premises, sovereign cloud and air-gapped use; customers bring their own model, with Nemotron or Poolside Laguna for full isolation; hybrid mode routes selected workloads to Claude, Gemini or GPT models; and Premium Packages extend Bob to Java, IBM i and IBM Z modernization.