@unihodl/agent-sdk

The TypeScript SDK for the UNIHODL Agent Handoff SDK — the decision-continuity protocol for human-to-agent and agent-to-agent work transfer.

Pass a UNIHODL resume token to any AI agent. The agent receives the human's open tabs, scroll positions, video timestamps, AI-tagged decision thread, partial conclusions, and intended next step — not just a list of links. It can pick up the work mid-thought.

Spec: https://unihodl.app/sdk/spec Roadmap & open gaps: https://unihodl.app/sdk/roadmap

Install

npm install @unihodl/agent-sdk
# or
pnpm add @unihodl/agent-sdk

Quickstart

import { Client } from "@unihodl/agent-sdk";
import Anthropic from "@anthropic-ai/sdk";

const uh = new Client({ apiKey: process.env.UNIHODL_API_KEY! });
const claude = new Anthropic();

// Hydrate the session into a structured Resume Context.
const ctx = await uh.sessions.hydrate("ses_8f3aZ91b", {
  audience: "claude.anthropic.com",
  scopes: ["read:context", "read:reasoning"],
});

const resp = await claude.messages.create({
  model: "claude-sonnet-4-7",
  max_tokens: 2048,
  system: ctx.asSystemPrompt(),
  messages: [
    { role: "user", content: "Continue Sarah's research on API v3." },
  ],
});

API

new Client({ apiKey, baseUrl?, fetch? })

Construct an SDK client. Send uh_test_* for sandbox or uh_live_* for production.

client.resume_tokens.create({ session_id, audience, scopes, ... })

Mint a scoped resume token bound to a session and audience. Returns the signed JWT.

client.sessions.hydrate(sessionId, opts?)

Hydrate a session into a HydratedContext. If opts.token is omitted, the SDK mints a one-shot token first.

HydratedContext.asSystemPrompt() renders the prompt-ready text block described in spec §6 — drop it directly into a Claude system prompt or a Gemini system_instruction.

Status

This is v0.1.0 — public preview matching spec v1.0. Known gaps are tracked publicly at https://unihodl.app/sdk/roadmap.

License

MIT — see LICENSE.