Gå til hovedinnhold

neokapi: Architecture

neokapi is an open content engine in Go. It reads supported formats into a shared content model, runs editing and checking tools, and writes content back through the format's writer. The kapi project tool, desktop app and agent interfaces use this engine. Applications can also import the Go library directly.

The engine separates content processing from the project's applicable terms, writing guidance and check policy. A check result names the analysis performed; a successful gate does not establish factual accuracy or other qualities that were not assessed. Multilingual content, AI ingestion and programmatic editing use the same format-aware processing interfaces.

Fidelity depends on the format and operation. No-op byte preservation, preservation outside an edit and semantic round-trip fidelity are distinct properties. Consult the format evidence for the applicable reader and writer. Start with the Go quickstart, or read the Architecture Decisions for the system's design.

Processing Pipeline​

neokapi processing architectureA streaming pipeline: a format reader at the left edge and a writer at the right that sinks to a file (round-trip) or a content store (overlays); a serial chain of annotate, translate and check tools in the middle, with the translate stage fanning out across parallel goroutines; content-memory, terms and a translation provider (LLM or MT) feeding it from above; and a gRPC plugin band (Okapi bridge, tier-3 segmenter/media/OCR plugins, remote plugins) feeding it from below. Each stage is a goroutine joined by Part channels, and the whole pipeline runs over many documents in parallel.documents in parallel · MaxConcurrencyTermsterm-lookupContent MemoryrecycleProviderLLM or MTSourcessource binding · fileapp.jsonpage.htmlguide.docxstrings.xmlReaderDataFormatchanAnnotatesegment · terms · NERchanfan-out · N goroutinesordered fan-inTranslategoroutine 1Translategoroutine 2Translategoroutine 3Translategoroutine 4chanChecksqa · enforcechanWriterDataFormatSinksink binding · file or storefileround-tripstoreoverlaysPlugin system · gRPC subprocess bridgeOkapi BridgeJava filtersTier-3 pluginssegmenter · media · OCRRemote / native plugincustom tool · format

The edges are the flow's source and sink: bindings that decide where content enters and leaves. The default, shown above, is the file binding: a reader turns supported source files into a stream of Parts and a writer turns the stream back into files with the fidelity its format supports. The same flow can instead bind to the project store, a .kpz workspace, or an interchange file, with no reader or writer (flows: source and sink). Between the edges runs a flow: a serial chain of tools connected by buffered channels of Parts. The tools divide by capability: annotators attach stand-off overlays and annotations (segmentation, terminology, entities, check findings, analysis results), translators fill in targets, and validators check and enforce, while content memory and the terms store feed the relevant stages.

Concurrency runs at three levels at once: each stage is its own goroutine joined by channels with automatic backpressure; a block-handling stage such as translation can fan out across N goroutines with an ordered fan-in; and the executor runs many documents in parallel, bounded by MaxConcurrency. Context cancellation propagates to every stage. Readers, writers, and tools can be supplied by plugins (the kapi-sat segmenter, the kapi-pdfium PDF reader, or any remote plugin), dispatched as subprocesses over gRPC. See F-01 and E-01.

Package Layout​

The tree below is trimmed to the packages the rest of these pages refer to; ls core in the repository lists the rest.

neokapi/
├── go.mod                           # module github.com/neokapi/neokapi
├── go.work                          # coordinates the framework, host, CLI, app and plugin modules
│
├── core/                            # Platform-agnostic framework packages
│   ├── model/                       # Part, Block, Layer, Run, Target, Overlay, Anchor, Data, Media
│   ├── format/                      # DataFormatReader/Writer interfaces, detection
│   ├── formats/                     # Built-in format implementations, one package each
│   ├── tool/                        # Tool interface, BaseTool dispatch, EditPlan
│   ├── tools/                       # Built-in tools (pseudo-translate, qa, segmentation, …)
│   ├── flow/                        # Executor, Builder, FlowDefinition, placement pass
│   ├── registry/                    # FormatRegistry, ToolRegistry
│   ├── ai/                          # LLM-backed tools and prompt assembly
│   ├── check/                       # Report and Finding
│   ├── project/                     # kapi.yaml recipe, context axes, gates
│   ├── profile/                     # Voice profiles and term rules
│   ├── kbf/                         # Content bundle (.kbf.json) and its validator
│   ├── plugin/                      # Plugin manifests and the gRPC protocol
│   └── …                            # segmentation, redaction, graph, state, storage, and the rest
│
├── memory/                          # Content memory
├── terms/                           # Terms store
├── kpz/                             # Project archive (.kpz)
├── providers/                       # LLM provider backends
│
├── host/                            # Cobra-free runtime + services (module: …/host)
├── cli/                             # Thin Cobra shell over host (module: …/cli)
├── kapi/                            # The kapi binary (module: …/kapi)
├── apps/kapi-desktop/               # Kapi Desktop (Wails v3; module: …/kapi-desktop)
├── plugins/                         # In-repo plugins, each its own module
├── packages/                        # Shared TypeScript packages (@neokapi/ui-primitives, @neokapi/flow-editor, …)
└── docs/                            # Repo internals (format ops, testing, runbooks)

The framework module (repo root) stays platform-agnostic. memory/, terms/, kpz/ and providers/ are top-level framework packages rather than nested under core/. Front-ends such as the CLI and the desktop app, and any other consumer, attach through the plugin and extension registries rather than by direct imports, so the framework never depends on a particular platform.

The framework concepts​

To see these concepts working together in a few lines of Go (register the formats, read a file into the content model, run a tool, and write the result), start with the Go quickstart. The framework rests on a few concepts, each with its own page:

  • Content Model: the format-independent representation. A document becomes a stream of Parts carrying layers, blocks, runs, overlays, data, and media. Embedded content (HTML inside JSON, CDATA in XML) is modeled as nested layers, each with its own format.
  • Formats: paired readers and writers that produce and consume the content model.
  • Tools: the processing units. Each reads Parts from a channel, transforms them, and writes them out.
  • Flows: named, ordered compositions of tools.
  • Pipeline: the concurrent executor that runs a flow: goroutines, buffered channels, and context-driven cancellation.

For the concrete Go interfaces and method signatures behind these concepts, see the Interface Reference. For the design rationale, see the Architecture Decisions.