@ai-matrx/content-ir
The framework-independent Content IR kernel used by AI Matrx applications. It parses streamed or complete structured content into canonical envelopes, manages incremental parse sessions, converts kind schemas, and performs structural validation.
pnpm add @ai-matrx/content-ir
Entry points
The root export is the full surface and always will be. Since 0.10.0 each internal area is also addressable on its own — prefer these, so a consumer that only reads wire contracts never pulls the parser (or ajv) into its graph.
| Import | What it holds |
|---|---|
@ai-matrx/content-ir | everything below |
@ai-matrx/content-ir/core | IR types + tree, JSON tokenizer, kind parser, snapshots, normalize, fingerprint, discriminator, envelope read/value/cache, region, schema structure |
@ai-matrx/content-ir/session | ParseSession and the one-writer-per-identity registry |
@ai-matrx/content-ir/registry | kind registry types, storage transform, the dual gate, createKindValidator over a host SchemaSourcePort |
@ai-matrx/content-ir/convert | kind ⇄ JSON Schema, the OpenAI structured-output converter |
@ai-matrx/content-ir/wire | partial-kind events, runtime wrapper readers, emit payload |
@ai-matrx/content-ir/directives | Kind Directives: slug grammar, the ONE decoder |
@ai-matrx/content-ir/source | offset-preserving source tokenizer, splice save + anchor-mapping change set, the single-dollar math rule |
import { createKindStreamParser, envelopeFromCompleteValue } from "@ai-matrx/content-ir/core";
import { validateStructuralLeg } from "@ai-matrx/content-ir/registry";
import {
readPartialKindEvent,
makePartialKindStalenessGate,
rehydrateNodeOutcome,
rehydrateRunResult,
withRootKind,
} from "@ai-matrx/content-ir/wire";
Reading an envelope: kindState, never kind alone
root.kind is preserved when a payload fails its schema — a broken flashcard_set stays a flashcard_set so you can say what broke. root.status reports only that parsing finished. The validity signal is root.kindState:
"resolved"— the only state a typed consumer may act on."unverified"— the kind has no registered schema, so nothing was ever checked; it still routes to its component."raw"/"pending_*"/"speculative"— do not consume as typed data.
The package has no browser, React runtime, application store, database client, routing, or renderer dependency. ajv is its only runtime dependency. Host-specific registry fetching, component compilation, rendering, and error capture remain adapters around this kernel.
/source streaming mode: what "settled" promises
tokenizeSource(text) reads stored text; tokenizeSource(prefix, { streaming: true })
reads a message still arriving. In streaming mode anything the next bytes could
still change is held back as plain prose (an open last line, an unclosed code
span, a $$ that can still pair, a footnote whose continuation is undecided,
lines a box-drawing tree still forming below may claim as its root label, a
paragraph cut short by a block that may still dissolve, …),
so an island reported complete is the island the finished text has — with
ONE documented exception:
- A late orphan reasoning closer. Models do emit closer-only reasoning (the
<thinking>opener lost upstream). When a</thinking>/</think>/</reasoning>arrives with no opener, the one-shot reading folds everything before it into a thinking region — retroactively. No streaming reader can know that in advance short of holding ALL prose provisional until the stream ends, which costs more than it saves (chair ruling, 2026-09-26; common-docsprojects/rich-content-unification/REGISTER.md, RC-B3 rulings).
Consumer contract: when a stream completes, re-read the finished text in
one-shot mode (tokenizeSource(text), no streaming) and show THAT — the
final screen always equals the one-shot result, which is also what a reload
shows. Stored text never goes through streaming mode, so none of this can
change a stored byte. (matrx-frontend's chat renderer does the same for its
own live accumulator: components/mardown-display/chat-markdown/settle-stream-blocks.ts.)
Release gate
pnpm --filter @ai-matrx/content-ir typecheck
pnpm --filter @ai-matrx/content-ir test
pnpm --filter @ai-matrx/content-ir check:package
The gate runs the parser's shared fixtures and parity vectors, validates the packed declarations and exports, installs the tarball in an empty project, and imports the public API.