Quick Start

Install

TypeScript

pnpm add @nekte/client @nekte/server

Python

pip install nekte

5-Line Quickstart

TypeScript

import { NekteClient } from '@nekte/client';
const client = new NekteClient('http://localhost:4001');
const catalog = await client.catalog();          // L0: ~8 tok/cap
const result = await client.invoke('sentiment', { input: { text: 'Great!' } });
console.log(result.out);                         // "positive 0.9"

Python

from nekte import NekteClient
client = NekteClient("http://localhost:4001")
catalog = await client.catalog()                 # L0: ~8 tok/cap
result = await client.invoke("sentiment", input={"text": "Great!"})
print(result.out)                                # "positive 0.9"

What Just Happened?

1. catalog() sent a nekte.discover request at Level 0. The server responded with a list of capability names, categories, and version hashes — roughly 8 tokens per capability instead of 120+ for full schemas.

2. invoke() sent a nekte.invoke request with the version hash from the catalog. Because the hash matched, the server executed the capability without requiring the schema to be re-sent — zero extra tokens.

3. The result came back in minimal format by default: a short string representation that fits in a few tokens.

Core Concepts

Concept	What It Means
L0/L1/L2	Discovery levels: catalog (8 tok) to summary (40 tok) to full schema (120 tok)
Version hash	8-char hash of a capability’s contract. If unchanged, skip schema reload
Token budget	{ max_tokens, detail_level } — the receiver adapts response granularity
Multi-level result	Same data in minimal/compact/full representations
DelegateStream	{ events, cancel(), taskId } — streaming with lifecycle control

Next Steps

• TypeScript Setup Guide — Full walkthrough: server, client, streaming, gRPC, bridge
• Protocol Overview — Why NEKTE exists and how it compares to MCP and A2A
• Primitives Reference — All 8 protocol primitives with JSON examples
• Architecture — Hexagonal architecture and DDD patterns