@belto_inc/slyos

Belto SlyOS SDK for local-first and hybrid AI infrastructure.

npm install @belto_inc/slyos

import { Belto } from "@belto_inc/slyos";

const belto = new Belto({
  apiKey: process.env.BELTO_API_KEY!,
  mode: "local-first"
});

await belto.initialize();

const response = await belto.chat({
  messages: [{ role: "user", content: "Summarize this document" }],
  rag: true
});

console.log(response.text);
console.log(response.route);

By default the SDK runs in local-first mode and auto-detects real local runtimes:

• Node/Electron/dev machines: AutoLocalRuntimeAdapter uses Ollama when http://127.0.0.1:11434 is available.
• Browsers: AutoLocalRuntimeAdapter uses WebLLM when WebGPU and @mlc-ai/web-llm are available.
• Cloud fallback: SmartRoute uses the Belto API and Amazon Bedrock only when local is unavailable, blocked by policy, or the request is better suited to cloud.

For the simplest local test:

ollama serve
ollama pull llama3.2:3b

Then run the SDK with mode: "local-first". SmartRoute will report a local route for simple prompts when the local model is available.

Edge RAG is also enabled by default:

await belto.addDocuments({
  storageStrategy: "edge-first",
  maxEdgeStorageBytes: 25 * 1024 * 1024,
  cloudSpillover: true,
  knowledgeBaseId: "kb_optional_for_cloud_spillover",
  documents: [{ filename: "faq.txt", rawText: "Belto keeps this context on device first." }]
});

Node/Electron stores edge chunks under ~/.belto-slyos/rag; browsers store them in IndexedDB.