Client SDK

Cognipeer Client SDK

The official JavaScript/TypeScript SDK for building conversational AI applications, with client-side tool execution — your functions run in your environment, and the results flow straight back into the conversation.

• Package @cognipeer/sdk
• Runtime Node.js · browsers · edge, ESM & CommonJS
• Tool calling OpenAI-compatible function format

Unlike SDKs that only support server-side tools, the Client SDK lets your JavaScript functions be called automatically by the AI, with full type safety and configurable retry logic — so sensitive operations stay on your own infrastructure instead of round-tripping through someone else's server.

npm

npm install @cognipeer/sdk

yarn

yarn add @cognipeer/sdk

pnpm

pnpm add @cognipeer/sdk

import { CognipeerClient } from '@cognipeer/sdk';

const client = new CognipeerClient({
  token: 'pat_your-personal-access-token',
  hookId: 'your-channel-hook-id'
});

const response = await client.conversations.create({
  messages: [{ role: 'user', content: 'Hello! How can you help me?' }]
});

console.log(response.content);

Guide

Getting started

Install the SDK, authenticate, and send your first message.

Guide

Client-side tools

Define functions the AI calls automatically, with typed arguments and results.

Guide

Flows

Run structured workflows and automations with typed input and output.

Reference

API reference

CognipeerClient, conversations, flows, and every exported type.

Examples

Examples

Basic usage, client tools, structured output, browser integration, and real-world use cases.

Project

Release notes

What shipped, release by release.

---

Why the Client SDK

Automatic tool execution

No need to manually handle tool calls — the SDK detects them, runs your function, and feeds the result back into the conversation for you.

Client-side security

Sensitive operations run in your own environment. Nothing needs to be proxied through a server just to call a function.

OpenAI-compatible tool format

Tool definitions use the same function-calling shape as OpenAI, so existing schemas and mental models carry over directly.

Universal runtime support

Works in Node.js, browsers, and edge runtimes, shipped as both ESM and CommonJS with full TypeScript types.

Where to go next

• Getting started — install, authenticate, first conversation.
• Client-side tools — define and register functions the AI can call.
• Configuration — client options, retries, and error handling.
• TypeScript support — types for messages, tools, and responses.