Napcar

Common Napcar integration patterns — detection, streaming UI, structured extraction, and embeddings.

Feature-detect and fall back

import { createClient, isNapcar } from "@napcar/sdk";

if (!isNapcar()) {
  showInstallBanner(); // degrade gracefully — never crash
}
const client = createClient();

Local in Napcar, your cloud provider elsewhere

One call site, two backends. In Napcar the prompt runs on the local model; in any other browser it routes to the cloud provider you configure. The call_llm("…") pattern is the one-shot prompt() helper:

import { prompt } from "@napcar/sdk";

const text = await prompt("Summarize this page", {
  // Used only when NOT in Napcar (no local binding). Ignored in Napcar.
  cloud: { provider: "openai", apiKey: OPENAI_API_KEY },
});

Or keep an explicit client and reuse sessions — same routing:

import { createClient } from "@napcar/sdk";

const client = createClient({
  cloud: { provider: "groq", apiKey: GROQ_API_KEY }, // openai | openrouter | groq | together | mistral | deepseek | ollama | custom
});
const session = await client.requestSession({ task: "chat" });
const { text, model } = await session.generate({ input: "Hello" });
console.log("local:", client.isNative, "model:", model);

A custom provider points at any OpenAI-compatible endpoint:

createClient({
  cloud: { provider: "custom", baseUrl: "https://api.example/v1", model: "my-model", apiKey },
});

Streaming into the UI

const session = await client.requestSession({ task: "chat", localOnly: true });
const output = document.querySelector("#out");
for await (const chunk of session.generateStreaming({ input: prompt })) {
  output.textContent += chunk;
}

Structured extraction

const { text } = await session.generate({
  input: invoiceText,
  responseFormat: {
    type: "json_schema",
    schema: {
      type: "object",
      properties: { vendor: { type: "string" }, total: { type: "number" } },
      required: ["vendor", "total"],
    },
  },
});
const invoice = JSON.parse(text);

Embeddings for local search

const { embeddings } = await client.embed({ input: docs });
// build a local vector index — nothing leaves the machine

OpenAI-compatible client

const res = await fetch("https://local.napcar.ai/v1/chat/completions", {
  method: "POST",
  headers: { "Content-Type": "application/json", "X-Napcar-LLM-Handoff": "optional" },
  body: JSON.stringify({ model: "napcar-default", messages, stream: true }),
});

Examples