Open the closed
Workflow engine, AI gateway, plugin system — running on your machine. Swap any component with a config line.
Every vendor locks you in gradually. It starts with a smooth onboarding, competitive pricing, great integrations — until the day switching costs more than staying. That's not a technical problem. It's a business model.
The second problem is less obvious. Large companies have entire engineering teams building internal developer platforms — AI orchestration, workflow engines, API gateways, observability stacks. All of that makes everything else faster. Small teams don't have that. They piece together SaaS subscriptions, each one owning another slice of the stack, each one extracting rent.
KB Labs is the answer to both. Not another SaaS layer on top of AI APIs — an infrastructure layer you install, configure, and own. One that was designed from the start to be replaced at every boundary without a migration project.
Open the closed is not a tagline. It's a design requirement: every system boundary must be open for inspection, replacement, and extension. The concrete implementation is always a detail. The typed contract is what lasts.
How we build
The boundary is enforced by the build
A plugin cannot import an adapter — the build fails. That's not a convention or a code review note: adding `@kb-labs/adapter-redis` to a plugin's dependencies violates the layer graph and breaks the build. The contract is the only connection point.
Your code. Your data.
KB Labs runs on your infrastructure. No cloud backend owned by a vendor, no data leaving your perimeter unless you explicitly route it. On-prem and air-gapped by design.
12 contracts across the platform
ILLM, ICache, IStorage, IVectorStore, IEventBus, ILogger — and six more. Each describes a capability, not a provider. All 12 converge in a single PlatformContainer — the only gateway through which plugins access infrastructure.
Level the playing field
Enterprise-grade internal developer platforms shouldn't require an enterprise budget. Any team — one person or a hundred — should be able to run AI orchestration, workflow automation, and a proper API gateway.
Plugins, not features
Everything optional lives in a plugin. No base classes, no framework inheritance — just a typed contract and a manifest file. If you can write a module, you can write a plugin.
Observability by default
Software you cannot inspect is software you cannot trust. Every workflow run, every agent call, every state mutation is traceable without attaching a debugger or adding extra tooling.
In 2025, the CTO at my company rolled out AI code review for the backend team. As the frontend lead, I disagreed — with both the tool and the approach. I built an MVP AI Review for frontend and backend together overnight. We discussed the architecture the next day; half my proposals made it in. I wanted more.
I published the first version on GitHub as open source. Then I wanted something bigger — not just a review tool, but real infrastructure for AI tooling. Step by step it grew: workflow engine, plugin system, AI gateway.
Today KB Labs is 145+ packages, its own release pipeline, and Go binaries for building and managing services. Public release is summer 2026. Source is open — `@kb-labs/review` is one of the first plugins.
Build on something that stays open.
Install KB Labs on your own machine, run it on one real workflow, and keep full architectural freedom as you grow.