UnisonQuick setup examples
Node.js Quickstart
Call the Unison brain from Node.js using the built-in fetch API — no extra dependencies required.
Node 18+ ships with fetch globally, so no extra HTTP library is needed. For a typed, higher-level client consider the official SDK instead: TypeScript SDK.
Prerequisites
export UNISON_TOKEN=usk_live_...Node 18 or later (for native fetch).
Recall — fetch prompt-ready context
const BASE = "https://brain.unisonlabs.ai";
const headers = {
Authorization: `Bearer ${process.env.UNISON_TOKEN}`,
};
const params = new URLSearchParams({ q: "payment service architecture", k: "5", mode: "auto" });
const res = await fetch(`${BASE}/v1/brain/context?${params}`, { headers });
const data = await res.json();
const memory = data.weakEvidence ? "" : data.contextMd;
// Prepend `memory` to your system prompt before calling the LLMPersist — save a conversation to the brain
const res = await fetch(`${BASE}/v1/brain/remember`, {
method: "POST",
headers: { ...headers, "Content-Type": "application/json" },
body: JSON.stringify({
dump: {
turns: [
{ role: "user", content: "What queue library should we use?" },
{ role: "assistant", content: "Switched to pgmq — simpler ops than Redis." },
],
},
sourceRef: "session-1",
}),
});
const { jobId } = await res.json();
console.log("jobId:", jobId);remember runs the save-or-skip curation pipeline (dedupe, curated notes, entity facts) as a background job. The returned jobId tracks it — poll GET /v1/brain/jobs/{jobId} for status. New memories surface in recall queries within seconds.
To write a document directly instead (no curation, explicit path/tags), POST a { type: "document", content, ... } item to /v1/brain/ingest — see Ingest.