Architecture Portal

See what works now and what remains to connect

Working CREXX, DSLSH and THE components cover language, execution and editor responsibilities. crexx-rag and Cognitive Pipelines contain current software for separate retrieval and desktop-orchestration roles. The new CoreLang language, Open BPM reference component and every cross-project composition remain proposed.

Start with working language, execution and editor components

CREXX contains the cREXX compiler, the RXAS assembler, RXLINK, CREXX virtual machines, libraries and packaging tools. A cREXX program can therefore move from source, through assembly and linking, to bytecode execution. Within this portfolio, cREXX is primarily a surface for business rules, transformations and scripting. Other components remain free to use Rust, TypeScript, React or another suitable technology.

DSLSH receives document changes from an editor, keeps the corresponding parser state up to date and returns structured language information. THE CREXX Edition uses DSLSH and CREXX in a working editor and can produce parser-backed HTML and TeX source rendering.

THE also has an opt-in Preact and TypeScript browser interface. The proof of concept keeps the native editor as the source of truth while the browser renders semantic snapshots and returns typed actions, text and keys. Its current limits include a fixed 24-by-80 virtual screen, full-state snapshots, one controlling client, no browser-exposed macros or transient dialogs, and a deployment model that leaves authentication and session routing to a reverse proxy. It is a working experiment, not a production multi-user web editor.

These working components give later experiments a concrete starting point. They do not yet supply the new CoreLang language, parser, common base or complete DomainLang workbench. A suitable future CoreLang or DomainLang example could emit RXAS and use the existing CREXX execution stack without adopting REXX language syntax or behaviour. That connection remains proposed. The LLVM compiler infrastructure remains later backend research.

Keep retrieval and desktop orchestration separate

crexx-rag is developing the knowledge, retrieval and source-grounding role. Its retained native executable keeps source passages, proposed and accepted claims, ambiguity and retrieval results distinct. Accepted cREXX slices cover source revisions, claims, source-backed packets, exact vector ordering, provider contracts and durable jobs. A schema-v2 storage foundation also exists.

The target cREXX product is not yet a complete ingestion and query path and has not replaced the native reference. The representative corpus is a capacity and migration fixture, not a promise of general production throughput, high availability or multi-worker operation.

Cognitive Pipelines is a working experimental Qt desktop application. A user can compose a graph with model calls, retrieval, scripts, external processes, control flow and human input, and then observe asynchronous execution. It is an AI and tool orchestration workbench, not a durable business-workflow service, distributed scheduler or production multi-user platform.

The next proposed slice connects distinct roles

A proposed composition would request source-backed material from crexx-rag, run defined computational steps in Cognitive Pipelines and ask an Open BPM interface to create durable human work. None of those integrations exists today. Open BPM’s reference component, application programming interface and participant interface have not been built, and Public Purpose Lab has not published an integrated demonstration. Architecture Portal owns the logical roles and quality questions; each project repository would remain authoritative for the software it implements.

Use rebooking to test one complete path

The rebooking scenario is deliberately small. It tests whether the method and blueprint can connect an authoritative appointment record, one repeatable work request, a versioned deadline rule, communication failure and recovery.

Open BPM could manage the work item. A cREXX rule could return the deadline or priority. Public Purpose Lab could introduce delays and failures with synthetic data and report what happened. Those connections have not yet been implemented, but the slice is concrete enough to challenge both the loop settings and the component catalogue.

Return each result through a delivery record

A useful record begins with the service question, the active loop settings, the alternatives considered and the person with decision authority. It links the changed code and configuration to defined checks, review decisions, operating observations and limitations. It ends by saying whether the lesson changes the implementation, an asset, the method or the blueprint.

For agent-assisted work, record what the agent changed, what a person reviewed and which actions were outside the agent’s authority. Then use the project map to follow each capability claim to its canonical source.

Diagram

Trace one question into operation and shared learning

A delivery record connects the initial uncertainty to an authorised decision, a small implementation, defined checks, operating observations and any change to shared guidance.

  1. Service question

    Name the affected people, intended result, constraint and uncertainty that the slice must resolve.

  2. Options and authority

    Compare realistic choices and name the person authorised to decide.

  3. Small implementation

    Build enough complete behaviour to test the important architecture decision.

  4. Checks and review

    Record the defined checks, observed results, human approvals and known limitations.

  5. Operation and recovery

    Observe real behaviour, support needs, failure and the actions used to restore the work.

  6. Returned learning

    Revise the implementation, asset, method or blueprint when the result justifies it.

Read the connections
  • Service questionOptions and authority: the uncertainty frames a decision.
  • Options and authoritySmall implementation: the authorised choice becomes testable behaviour.
  • Small implementationChecks and review: reviewers compare behaviour with the decision.
  • Checks and reviewOperation and recovery: an accepted slice enters realistic conditions.
  • Operation and recoveryReturned learning: observed use changes local or shared guidance.
Another reader should be able to trace a claim to the decision, changed implementation, observed result and limitation that support it.

Related work

What the other projects contribute

  • Each implementation repository owns its detailed capability, test and limitation claims.
  • Public Purpose Lab may own future integrated demonstration results.