@c0-ui/server

Server-side runtime for c0 — the open-source generative UI pipeline. System prompts, an OpenAI-compatible stream adapter, and a response builder. BYOK (Bring Your Own Key) — works with any OpenAI-compatible LLM, including local models.

Install

npm install @c0-ui/server
# openai is an optional peer dependency
npm install openai

Quick Start

import { createSystemPrompt, transformOpenAIStream } from '@c0-ui/server';
import OpenAI from 'openai';

const openai = new OpenAI(); // or any OpenAI-compatible endpoint (BYOK)

const stream = await openai.chat.completions.create({
  model: process.env.LLM_MODEL!,
  stream: true,
  messages: [
    { role: 'system', content: createSystemPrompt({ modelTier: 'weak' }) },
    { role: 'user', content: 'Show me last quarter sales as a table' },
  ],
});

// Returns a ReadableStream<string> of the c0 XML-DSL — pipe to the client
return new Response(transformOpenAIStream(stream));

Key exports

| Export | Purpose | |--------|---------| | createSystemPrompt(options) | Build the system prompt; modelTier: 'weak' tunes it for small models (7B–13B) | | transformOpenAIStream(stream) | Adapt an OpenAI chat stream into the c0 DSL ReadableStream<string> | | toOpenAIMessages(...) | Convert c0 thread state into OpenAI message format | | makeC0Response() | Manually build a streamed response (writeContent / writeArtifact / writeThink) | | BASE_PROMPT, C0_SYSTEM_PROMPT, DEFAULT_ARTIFACT_TYPES, MODEL_COMPATIBILITY | Prompt + artifact-type building blocks |

Why c0 server (vs a closed Gen UI API)?

The entire pipeline — LLM routing, the component-selection prompt, quality filtering — is open and runs in your process. No black-box server, no model lock-in, debuggable in production.

License

MIT © domuk-k