Development packages
Two ways to build.
One engineering standard.
Both packages deliver the same expert-led, AI-assisted engineering. They differ in where AI tools process your code and project context while we build, and in what has to be set up and run.
Software delivery with suitable privately hosted models, inside a processing environment we agree with you. It's an engineering engagement, not GPU rental: we select and evaluate models, set up development agents and their permissions, and deliver with the same expert design, QA and release control.
Best fit
Policies that keep source code, prompts and research inside a boundary you control.
- AI processing environment
- Open-weight models on your infrastructure or in an agreed isolated environment.
- Setup
- Model evaluation, environment and agent setup before development, scoped per engagement.
- Infrastructure responsibility
- Model hosting is part of the engagement, run by your team or ours as agreed.
- Ongoing management
- Optional ongoing model and infrastructure management.
Expert-led delivery using Claude Code and/or OpenAI Codex with approved cloud-model configurations. You buy our engineering capability, not a resold AI subscription.
Best fit
Teams that can approve commercial AI coding tools and want an AI-first workflow without hosting models.
- AI processing environment
- Claude Code and/or OpenAI Codex, processed by the provider under account settings agreed with you.
- Setup
- Accounts, repository access, data handling and retention settings agreed before development.
- Infrastructure responsibility
- No dedicated model infrastructure for this delivery approach.
- Ongoing management
- Tool administration and access reviews; product operations are scoped separately.
Both packages include
- Experienced engineering oversight
- Product design, as scoped
- Professional QA
- Security-conscious reviews
- Controlled releases
- Documentation
- Agreed handover
How the two packages compare
| Consideration | Private / Local AI Engineering | Claude Code / OpenAI Codex Engineering |
|---|---|---|
| Best fit | Policies that keep code, data and research inside a defined boundary | Teams that can approve commercial AI coding tools |
| AI processing environment | Privately hosted open-weight models on your infrastructure or an agreed isolated environment | Claude Code and/or OpenAI Codex, processed by the provider under agreed account settings |
| Setup | Model evaluation, environment and agent setup before development, scoped per engagement | Accounts, repository access and data settings agreed before development |
| Infrastructure responsibility | Model hosting is part of the engagement, run by your team or ours as agreed | None for models: the provider runs them |
| Ongoing management | Optional: model updates, monitoring, capacity and access reviews | Tool administration and access reviews; product operations scoped separately |
| Shared quality standards | Engineering oversight, design, QA, security-conscious reviews, controlled releases, documentation and handover | Engineering oversight, design, QA, security-conscious reviews, controlled releases, documentation and handover |
FAQ
Questions about the packages
How do we choose between the packages?
Start from your policies. If code and project context must stay inside a boundary you control, choose Private / Local AI Engineering. If your organization approves Claude Code or OpenAI Codex, Claude Code / OpenAI Codex Engineering avoids running model infrastructure. We can recommend one during scoping.
Do the packages decide where our product is hosted?
No. They describe the AI tools used while we build. Hosting for your finished product follows its own requirements and can be anywhere, including the cloud or your own servers.
Is the quality different between the packages?
No. Engineering oversight, design, QA, security-conscious reviews, controlled releases, documentation and handover are the same. What differs is the tooling environment and its setup.
Do we need a package for QA, DevOps or UX-only work?
Only if AI tools will process your code, data or research during that work. Standalone engagements simply follow whichever environment your policies allow.
Can we start with one package and switch later?
Yes. Switching mainly changes tooling and setup; the delivery process, reviews and standards stay the same.
Not sure which package fits?
Tell us about your project and any rules on where code and data may be processed. We'll recommend a package during scoping.