What If the App Was a Prompt?

By The Baker Street Team

Every few years, a shift in computing rewrites the rules. The mainframe gave way to the PC. The PC gave way to the web. The web gave way to mobile. Each transition changed not just where software ran, but how we thought about building it.

We believe the next shift is already underway. The application is becoming a prompt. Instead of clicking through menus and filling out forms, people describe what they want in natural language and an AI agent figures out the rest. The interface disappears. The intent is the interface.

But here is the problem: most AI agent platforms today are built like toys. They run as a single process on someone else's server. Your data leaves your network. Your prompts are logged by a third party. There is no isolation between agents, no audit trail you control, and no way to enforce the security policies your organization already has in place.

We built Baker Street because we believe AI agents deserve real infrastructure.

Why We Built Baker Street

The idea started with a simple question: what would an AI agent platform look like if it were designed by infrastructure engineers instead of demo builders?

The answer turned out to be Kubernetes. Not because Kubernetes is trendy, but because it already solves the hard problems that agent platforms are only beginning to encounter. Process isolation. Resource limits. Network policies. Role-based access control. Secrets management. Horizontal scaling. Health checks. Rolling deployments. These are not features you bolt on after launch. They are the foundation you build on from day one.

Baker Street is a Kubernetes-native AI agent platform. Every agent runs in its own pod with its own namespace, its own network policy, and its own resource constraints. There is no shared memory space. There is no way for one agent to read another agent's data unless you explicitly grant access through Kubernetes RBAC.

This is not security theater. This is the same isolation model that banks, hospitals, and defense contractors use to run their most sensitive workloads.

What Makes Baker Street Different

Most AI agent tools fall into one of two categories. The first is the hosted SaaS platform: easy to set up, impossible to audit, and a non-starter for any organization with real compliance requirements. The second is the open-source framework: flexible but unfinished, leaving you to solve deployment, security, and operations on your own.

Baker Street is neither. It is a complete, self-hosted platform that you deploy to your own Kubernetes cluster. Here is what that means in practice:

Self-hosted by design. Your data never leaves your network. Your prompts, agent memory, and outputs stay on infrastructure you control. You choose the LLM provider, or run a local model. There is no phone-home telemetry, no usage-based pricing, and no vendor lock-in.

Real security, not checkbox security. Every agent container runs as non-root with a read-only filesystem, seccomp profiles, and all Linux capabilities dropped. Images are signed and verified before deployment. The audit trail is HMAC-chained and tamper-evident, designed to forward to your existing SIEM infrastructure. These controls map directly to SOC 2, NIST 800-53, and CIS benchmarks.

Extensible through sidecars. Need an agent that can query a database, call an API, or interact with a legacy system? Extensions run as sandboxed sidecar containers with scoped permissions, resource limits, and their own network policies. They cannot escape their sandbox. They cannot access other agents. They are exactly as powerful as you allow them to be.

Multi-channel from the start. Baker Street agents are not trapped inside a single chat window. They connect to Slack, Discord, Microsoft Teams, webhooks, and custom integrations through a unified channel abstraction. One agent can serve multiple channels simultaneously, or you can dedicate agents to specific teams and workflows.

The Core Capabilities

Baker Street ships with six core systems that work together:

Agent Runtime manages the lifecycle of AI agents as Kubernetes deployments, handling creation, configuration, scaling, health monitoring, and graceful shutdown.

Memory gives each agent persistent, encrypted storage for conversation history, learned context, and structured knowledge. Memory is scoped per-agent and survives restarts.

Jobs provide a way to run one-off or scheduled tasks as Kubernetes Jobs, with full logging, timeout handling, and retry logic. Think of them as cron for your agents.

Tasks enable agents to break complex work into trackable units, with status tracking, dependencies, and progress reporting that surfaces through whatever channel the agent is connected to.

Extensions are the integration layer. Each extension runs as a sidecar container with a well-defined API surface, letting agents interact with external systems without compromising the security boundary.

Channels abstract the communication layer, so agents can receive and respond to messages from any supported platform through a consistent interface.

What Comes Next

Baker Street is open source under the MIT license. The code is on GitHub, the documentation is on this site, and the community is forming around discussions and contributions.

We are building Baker Street in the open because we believe infrastructure this important should be inspectable, auditable, and improvable by the people who depend on it. We are not building a product you rent. We are building a platform you own.

If you have been looking for a way to run AI agents that meets your organization's security and compliance requirements, without giving up control of your data or locking yourself into a vendor, Baker Street was built for you.

Check out the documentation to get started, explore the features to see what is possible, or join the community to help shape what comes next.

The application is becoming a prompt. Baker Street is the infrastructure that makes it safe to let that happen.