Quickstart

Add InferKit to any page in about two minutes.

1. Get a publishable key

Sign up at app.synapticortex.dev → verify your email → API Keys → Create key. Copy the publishable key (ik_pub_live_…).

Use the publishable key (ik_pub_…) in the browser. Keep the secret key (ik_secret_…) on your server only — never ship it to the page.

2. Drop in the script

<script src="https://cdn.jsdelivr.net/npm/synapticortex-chat/dist/inferkit.js"></script>
<script>
  InferKit.init({ apiKey: 'ik_pub_live_…' });
</script>

That’s it. A chat bubble appears, reads the page, and answers questions about it. On a WebGPU browser it loads a small model locally (cached after first load); otherwise it falls back to remote inference (paid tiers).

3. (Optional) Configure

InferKit.init({
  apiKey: 'ik_pub_live_…',
  model: 'auto',              // 'auto' | 'nano' | 'small' | 'balanced' | 'quality'
  groundingMinMatches: 1,     // raise to make the on-page grounding stricter
  autoOpen: false,            // open the widget on load
  onReady:  ({ mode, tier }) => console.log('ready', mode, tier),
  onError:  (e) => console.error(`[${e.code}] ${e.message} (ref: ${e.reference})`),
});

Install via npm (bundlers / frameworks)

npm install synapticortex-chat

import { init } from 'synapticortex-chat';
init({ apiKey: 'ik_pub_live_…' });

Requirements & notes

Local mode needs a WebGPU browser (Chrome/Edge; Safari is rolling it out). Non-WebGPU visitors automatically use remote mode on paid tiers.
First local load downloads model weights once (a few hundred MB), then caches them in the browser for instant subsequent loads.
Domain lock: restrict a publishable key to your domains in the dashboard so a scraped key can’t be used elsewhere.

Next: API & SDK reference · Security & Privacy