@c0-ui/protocol

Framework-agnostic core of c0 — an open-source generative UI pipeline. Streaming XML-DSL parser, serializer, and JSON repair for turning an LLM's token stream into structured, renderable UI parts.

This package has zero UI framework dependencies — it's the shared core that @c0-ui/react (and future @c0-ui/vue, @c0-ui/svelte) build on.

Install

npm install @c0-ui/protocol

Quick Start

import { createStreamParser } from '@c0-ui/protocol';

const parser = createStreamParser({
  repairJson: true, // auto-repair broken JSON from weak/local models
  onContent: (part) => console.log('content', part),
  onArtifact: (part) => console.log('artifact', part),
});

// Feed the LLM token stream chunk by chunk
parser.write('<content>Here is your data</content>');
parser.write('<artifact type="Table">{"rows":[...]}</artifact>');

const response = parser.getResult(); // ParsedResponse

What it does

The LLM emits an XML-DSL: <content>, <artifact>, and <thinkitem> tags. createStreamParser() parses this incrementally (via htmlparser2) and produces an immutable ParsedResponse on every event (via immer), so it plugs straight into reactive rendering.

Key exports

| Export | Purpose | |--------|---------| | createStreamParser(options) | Incremental streaming parser with onContent / onArtifact / onThink callbacks | | parseResponse(text) | Parse a complete (non-streamed) response | | serializeResponse / extractContext | Serialize parsed state back to the DSL / extract context | | repairJson(text) | Standalone JSON repair for malformed model output | | wrapArtifact / wrapContent / wrapThinkItem / TAGS | Tag builders for producing the DSL server-side |

Why repairJson?

Unlike closed alternatives that assume a top-tier model, c0 targets any OpenAI-compatible LLM. Small models (7B–13B) frequently emit broken JSON inside artifacts; repairJson: true recovers it on the fly.

License

MIT © domuk-k