Architecture

Prologue is deliberately self-contained. Every other Cratis product either runs inside an Orleans silo (Chronicle) or is a library you reference from one (Arc, Components). Prologue is neither — it’s three ordinary containers (or a downloadable binary, for the Extractor) that read and write plain files and HTTP, so it can stand next to any system, Cratis-based or not, without pulling in a hosting model of its own.

The pieces

• Extractor — a reverse proxy plus a database and telemetry watcher, all in one process. It’s the only piece that has to run continuously, next to the system it’s watching.
• Receiver — a thin HTTP endpoint (POST /captures, POST /prologues/{id}/captures) that stores whatever the Extractor posts to it in MongoDB via Cratis.Prologue.Storage. It’s optional — the Extractor can write JSON files to a mounted folder instead, with no Receiver or MongoDB involved at all.
• Interpreter — a job, not a service, by default. It reads a folder of capture files, interprets them, and exits. A second mode keeps it running as a resumable HTTP session backed by MongoDB, for Studio’s interactive review flow.

The shared Cratis.Prologue.Contracts, Cratis.Prologue.Configuration, and Cratis.Prologue.Interpreter.Contracts NuGet packages give all three tools (and any consumer, like Studio) one canonical shape for captures, configuration, and the extraction result — nobody hand-parses another tool’s output format.

Two ways to run the Interpreter

Batch mode is what you get from the image by default, and what the Cratis CLI’s cratis prologue interpret drives — a folder of captures in, an ExtractionResult and a Screenplay .play file out, then the container exits. See Running the Interpreter.

Service mode hosts the same interpretation logic as a resumable session over HTTP, with session state persisted in MongoDB rather than kept in memory — so a long-running interpretation (one that’s waiting on you to answer an LLM’s clarifying question, for instance) survives the container being recycled. This is the mode Studio drives; if you’re not integrating with Studio, batch mode is what you want.

No dependency on Studio

Nothing here requires Studio to be running, and nothing in Studio requires Prologue — Studio simply happens to be a good consumer of the ExtractionResult/MongoDB shape when you want to review and reshape a model visually before running it. If you only want a .play file, the Extractor and the batch Interpreter are the whole pipeline; the Cratis CLI is the most convenient way to drive them.

Next

• Point Prologue at your system — wiring the Extractor up to something real.
• Reference — Configuration — every cratis-prologue.json property.