There is a pattern in enterprise software that has persisted for decades: make the product easy to adopt, then make it difficult to leave. Proprietary data formats, closed APIs, vendor-specific abstractions that quietly rewire how your organization thinks about its own systems. By the time switching costs become visible, they are already structural.
We built Kraken AI to work differently. Not because openness is a marketing position, but because we believe it produces better infrastructure. Platforms that are easy to leave are platforms that have to keep earning their place. That pressure is productive. It forces us to be better at the thing we actually do: orchestrating, securing, and governing AI agents at scale.
Architecture as the guarantee
Portability is not a feature you bolt on. It is a consequence of architectural decisions made early and maintained rigorously. Here is what that looks like in practice.
Agent definitions are declarative and standard. The Agent Definition Format is a structured YAML manifest that describes what an agent perceives, how it reasons, and what actions it can take. It is not a Kraken-specific abstraction. It is a portable specification. You can inspect it, version it, and run it outside our platform.
Connectors are MCP-native. We did not invent a proprietary connector protocol. The Connector SDK builds on the Model Context Protocol, the open standard that has become the universal interface between AI systems and external tools. Every MCP server in the ecosystem is automatically a Kraken connector. Our value is the governance, authentication, and health monitoring layer we provide on top, not the protocol itself.
Models are yours to choose. Every tier ships with BYOK: bring your own keys, connect your own model providers. Anthropic, OpenAI, Azure, open-source endpoints, private models on your own infrastructure. We do not intermediate model access, and we do not bill for tokens. Your relationship with your model providers is yours.
Data is yours to export. Agent definitions, execution traces, policy configurations, audit logs. All exportable. All in standard formats. We retain your data based on your tier for operational continuity, but ownership is never in question.
Deployment on your terms
Kraken runs wherever your infrastructure lives. Dedicated cloud instances are the baseline. VPC peering and private networking are available for teams that need isolation. On-premises and air-gapped deployments exist for regulated environments where data cannot leave a specific boundary.
This is not an enterprise upsell. It is a reflection of how production infrastructure should work. Your agents operate on sensitive business data, make real decisions, and interact with real systems. Where that happens should be your call.
The honest version
We are not pretending every component is open source. The control plane, the safety engine, the observability suite, the visual builder: these are proprietary. They are how we build a sustainable business around the platform. The open-source foundation (the Agent SDK, the ADF specification, the Connector SDK, the CLI) creates the ecosystem. The proprietary control plane is the product.
This split is intentional and transparent. We follow the open-core model that has been validated by companies like HashiCorp, GitLab, and Supabase. Open what creates the ecosystem. Monetize what creates business value. The result is a platform where customers are never locked in by the format of their data, the definition of their agents, or the protocol their connectors use.
Why this matters now
AI agents are about to become load-bearing infrastructure. They will manage inventory, execute trades, route logistics, and make decisions that directly affect operations. When that happens, the switching cost of agent infrastructure will become the most consequential vendor dependency most enterprises carry.
The companies building agent infrastructure today are setting the terms for the next decade of enterprise software. Some will build moats through lock-in. We would rather build them through the quality of what we ship.
We want our customers to stay because the platform is the best place to run their agents, not because leaving is expensive.
For details on data export, deployment options, and API access across tiers, see Pricing.