Summary

Bluetext is an opinionated platform for building enterprise-grade distributed systems. A system is declared in one place and in one form; its code, its configuration and its deployment objects are generated from that declaration; and the technology that fills each capability is already chosen. It is built for expert developers who want to work at AI speed while staying in control of what gets built.

What that is worth to whoever is paying for the system:

Enterprise-grade stops being a project of its own. Compliance evidence, least privilege, secret handling, production parity, rotation, provider independence — practices that normally each become a workstream are carried by constructs instead, and most of them cost nothing to follow (§3). The same work stops being paid for again in every system and every environment.

AI speed without unreviewed code. The reason to want AI-powered development is throughput; the reason to fear it is that nobody read the output. Here what the AI writes is the declaration, which is small enough for a person to approve, and what follows from it is generated or checked (§5, §8). The output grows without the review surface growing with it.

Reviews and audits become reading rather than investigation. Each kind of decision has one home, each instance of a kind has the same shape, the sets are enumerable, and a few arrangements cannot be built at all (§4). What an auditor asks for is read off the declaration rather than assembled, and what a security review has to cover is a list rather than a search.

The running bill is lower, and it stays negotiable. Rust workloads, one data engine where most systems run three or four, a broker with no coordinator beside it, static bytes served off the application path, and workloads packed onto shared nodes. Because the deployment objects are provider-independent, changing cloud or moving in-house is a target rather than a migration — which is what keeps a hosting price a conversation (§6).

Two things are licensed, and both are things nobody should build. The identity server that holds the session boundary, and the database engine. Nothing is paid for that generation could have produced (§6.3, §6.5).

The decisions made now stay reversible. Self-hosted or managed, one tier or another, a container in development and somebody else's service in production — each is a line in a file, and nothing that uses the component changes. A choice that can be undone cheaply is a different kind of choice from one that cannot.

And what it does not do. It does not make a system compliant; it removes the work of producing the evidence and keeping it true. It does not make authorization correct; it makes it findable and complete. Two capabilities recommended here are designed and not yet built, and §9 lists them beside everything else that is missing. Nothing in this document is claimed to work that does not.