Hi - I answer from the OpenSmartRoute documentation: routing, the API, plans and quotas, self-hosting. Ask away, or open a support ticket if you need a person.
Grounded in the docs - follow a source before acting on it.
IBM Bob now available for self-hosted and air-gapped environments - OpenSmartRoute
IBM Bob now available for self-hosted and air-gapped environments
IBM has made its agentic software development platform Bob available for on-premises, private clouds, and air-gapped networks. Customers can bring their own models and run them locally.
Key points
Self-hosted IBM Bob is generally available for enterprise use
Customers must source, license, and host supported models themselves
Supported models include Claude Sonnet 5.0, Gemini 3.7 Flash, GPT 5.6 Sol
Customers can run inference locally or route workloads externally
Why it matters: Running models locally improves security, control, and compliance for sensitive or isolated environments.
By OpenSmartRoute editorial · written through the router by llm-small
From MarkTechPost - “IBM Brings Bob to Self-Hosted and Air-Gapped Environments: Agentic Software Development Without Moving Your Code”
Introduction - IBM makes Bob available for self-hosted deployment
IBM has announced that its agentic software development platform, Bob, can now be run on-premises and in air-gapped environments. This means customers can host Bob within their own secure networks without needing to connect to the internet. The new deployment option is generally available, allowing enterprise users to install and operate Bob locally. This change addresses the needs of organizations that require strict security and data privacy.
Previously, IBM Bob was mainly offered as a cloud service. Now, it can be installed directly on a company's servers or private cloud. This gives organizations more control over their data and infrastructure. They can run Bob in environments that are isolated from the internet, such as secure government or financial networks. The move also supports hybrid setups, where some parts of the system are local and others are cloud-based.
What is IBM Bob - features and full lifecycle coverage
IBM Bob is a platform designed to help developers work with code using artificial intelligence. It covers the entire software development process, from understanding code to testing changes. Bob can analyze code, suggest improvements, plan work, and validate results. It aims to make software development faster and more accurate by automating routine tasks.
The platform includes several core capabilities. It offers an integrated development environment (IDE) for coding and debugging. It has a tool called BobShell for command-line work. Bob supports parallel tool calling, which allows multiple AI tools to work together. It also includes an agent harness that manages AI agents and their skills. These features help developers automate many parts of their workflow.
Bob is designed to work with different modes and skills. It can adapt to various tasks and user needs. The platform is flexible enough for different types of projects, from small scripts to large enterprise systems. It also supports optional premium packages for specialized modernization tasks, such as updating Java code or working with IBM i and IBM Z systems.
Mirror Particle raises capital to create an AI engine that simulates changing human motivations. The company rejects large language models in favor of a foundation model trained on longitudinal data.
Mistral released a preview of Mistral Large 4, a one-trillion-parameter model trained in Europe.
The new deployment option makes IBM Bob suitable for a range of secure environments. Customers can install Bob on their own servers or private cloud setups. This gives them full control over the hardware and software. It also allows organizations to keep sensitive data within their own network boundaries.
Air-gapped networks are isolated from the internet, often used by government or military agencies. IBM Bob can now run in these environments, which require strict security measures. Organizations can source, license, and host supported models locally. This setup prevents data from leaving the secure environment, reducing risk.
IBM supports hybrid configurations, where some parts of Bob run locally and others connect externally. For example, a bank might run core banking inference locally but route less sensitive tasks to external models. This flexibility helps organizations balance security with access to advanced AI models. They can choose where and how to process different workloads based on their security policies.
Model layer and inference - supported models and routing options
In IBM Bob, the model layer determines where inference runs. Inference is the process of generating outputs from a trained AI model. Bob does not include pre-installed models but allows customers to bring their own models from a supported list. These models can be hosted either inside the customer environment or externally.
The supported models are grouped based on deployment method. For external service use, models like Claude Sonnet 5.0, Claude Opus 4.8, Gemini 3.7 Flash, and OpenAI GPT 5.6 are available. These models are hosted by approved external providers and accessed via private connectivity patterns. This setup allows sensitive code and data to stay within the customer’s environment while still leveraging powerful external models.
Routing options enable organizations to decide where code and context are processed. For example, a bank could run inference locally for core applications. Less sensitive workloads could be routed to external models. This setup offers flexibility and control over data flow and security. IBM also plans to expand the model portfolio and introduce multi-model routing, which would allow switching between models based on workload needs.
Supported models - list and grouping by deployment method
The list of supported models includes several options for different deployment scenarios. For external service use, models like Claude Sonnet 5.0, Claude Opus 4.8, Gemini 3.7 Flash, and OpenAI GPT 5.6 are supported. These models are hosted by external providers and accessed through approved private connections.
For on-premises or air-gapped environments, organizations can bring their own models. Examples include Nemotron or Poolside Laguna, which are designed for full isolation. These models are hosted locally within the organization’s secure network. This approach ensures that sensitive data remains protected and does not leave the environment.
IBM also offers a hybrid mode. In this mode, some workloads are routed to external models like Claude, Gemini, or GPT. This allows organizations to use the best models for each task while maintaining security. Premium packages extend Bob’s capabilities to support modernization of legacy systems like Java, IBM i, and IBM Z.
Comparison with competitors - scope and target use cases
IBM Bob’s self-hosted version focuses on security and control. It is designed for organizations that need to keep their data within their own environment. Its main competitors include platforms like GitLab Duo Agent Platform, Mistral Vibe for Code, and GitHub Copilot CLI.
GitLab and GitHub provide agent-based development tools integrated into their own DevOps platforms. They mainly support cloud-based or internet-connected workflows. Mistral pairs its agent with its open-weight models, suitable for air-gapped use but limited to its own ecosystem.
IBM targets long-standing, enterprise systems that require modernization without exposing data to the internet. Its platform supports legacy environments like IBM i and IBM Z, making it suitable for industries with strict security needs. The focus is on integrating AI into existing, secure infrastructure rather than cloud-only solutions.
Why it matters - benefits for secure and isolated environments
The ability to run IBM Bob in self-hosted, air-gapped environments offers significant benefits. Organizations with strict security policies can now use AI-powered development tools without risking data leaks. They can keep sensitive code, customer data, and infrastructure details within their own networks.
This deployment flexibility supports industries like finance, government, and defense. These sectors often operate in highly secure environments and cannot connect to the public internet. IBM Bob allows them to modernize their workflows while maintaining compliance with security standards.
Using local models and routing workloads selectively enhances control. Organizations can decide which tasks stay inside their environment and which can leverage external AI services. This approach balances security with the need for advanced AI capabilities. It also simplifies compliance with data privacy regulations.
How it compares - what existed before, what this changes and what stays the same
Before IBM Bob offered only cloud-based deployment options. Customers had to run the platform on IBM’s cloud or public cloud providers. This limited use in environments with strict security or data privacy needs.
The new self-hosted option changes that. It allows enterprises to run Bob entirely on their own infrastructure. They can host it in private clouds, on-premises servers, or air-gapped networks. This gives organizations more control over their data and security.
The core capabilities of Bob remain the same. Users still get the IDE experience, BobShell, parallel tool calling, and the agent harness. These features support the full software development lifecycle. The self-hosted version also supports integration with supported IDEs and infrastructure environments.
What stays the same is the ability to bring your own models. Customers can license models from IBM’s supported list and run inference locally. They can route some workloads externally if needed. The platform still offers optional premium packages for specific modernization tasks.
However, the deployment scope expands significantly. Previously, organizations could only use Bob via IBM’s cloud. Now, they can run it in environments with no internet access or limited connectivity. This makes Bob suitable for industries with high security and compliance requirements.
The supported models at general availability include Claude Sonnet 5.0, Claude Opus 4.8, Gemini 3.7 Flash, and OpenAI GPT 5.6 Sol. These models can be hosted externally or internally, depending on the organization’s needs. The choice of where inference runs depends on the deployment method.
In summary, IBM Bob’s self-hosted deployment broadens its usability. It maintains its core features while adding deployment flexibility. This change targets organizations that need to keep their data within secure environments.
Questions this leaves open - what the source does not say and how a reader can check it
The source does not specify the exact hardware requirements for deploying Bob in self-hosted environments. It also does not detail the licensing process or costs involved. These are important for organizations planning to implement the platform.
It is unclear how the licensing works for different models. For example, whether licensing is per user, per model, or based on usage. The source mentions bring-your-own-license (BYOL) for supported models but does not explain the process or restrictions.
The source does not specify the supported operating systems or infrastructure platforms. It mentions integration with IDEs and infrastructure environments but does not list specific software or hardware requirements. Organizations should verify compatibility with their existing systems.
The details about model hosting are limited. It is not clear if organizations need to set up their own inference servers or if IBM provides any management tools. Clarifying this can help organizations plan their deployment architecture.
The source mentions that IBM plans to expand the model portfolio and add multi-model routing. It does not specify the timeline or which models might be added. Interested users can check IBM’s official documentation or contact IBM sales for updates.
It is also not clear how updates and maintenance are handled for self-hosted deployments. Do organizations receive regular updates? How are security patches applied? These are critical questions for secure environments.
Finally, the source does not specify the level of technical support available for self-hosted deployments. Organizations should inquire about support options, SLAs, and training resources from IBM.
To get answers, organizations can request a demo or contact IBM directly. They can also review IBM’s official documentation and licensing policies. Consulting with IBM sales or support teams can clarify deployment details and costs.
What to do - how to implement and get started
Organizations interested in deploying IBM Bob locally should start by sourcing supported models. They need to license and host these models within their environment. IBM recommends requesting a demo to understand how the platform integrates with existing infrastructure.
Next, they should evaluate their security policies and decide which workloads to run locally and which to route externally. Setting up the environment involves installing Bob on supported hardware or private cloud systems. Integration with existing IDEs and infrastructure is straightforward, according to IBM.
Once installed, users can configure routing options and connect to external models if needed. Organizations should test the setup thoroughly to ensure security and performance. IBM offers optional premium packages for specific modernization tasks, which can be added later.
Finally, organizations should train their teams on how to use Bob effectively. They can leverage IBM’s documentation and support services to optimize deployment. Regular updates and evaluations will help maintain security and performance standards.
