Self-hosted does not mean self-explanatory

There is a real appeal in keeping repositories and execution on machines you control. You can choose the environment, keep the existing toolchain and decide how it is maintained.

But the label self-hosted does not answer every important question. An agent can still call an external provider. A relay can still be part of the access path. A backup can exist without ever having been restored.

Draw the browser, the application service, the execution host, its data stores and the external services it talks to. For each connection, ask what crosses it and who controls it.

This exercise is useful even for a small deployment. It separates code execution from browser access, access from authorization and a retained conversation from a tested recovery mechanism.

A polished happy path is not enough for a workspace meant to hold long-running work. When visibility disappears, the product should help establish what happened before offering to repeat an action.

That matters because development commands can change state. Replaying an uncertain step may duplicate the effect, while restarting too eagerly may discard useful context.

Qodarium's intended boundary keeps code and compute on customer infrastructure while supporting services may handle access and product metadata. Exact behavior must be validated and documented with releases.

The practical goal is an operating model a small team can understand. Start with the host-readiness checklist and read the architecture guide for the product boundary.