Documentation

Everything you need to build AI agents on edge.

1Getting Started

Install

npm install @carloscortezcloud/tinkuy-agent @carloscortezcloud/styrr-llm

Add sayay-guard for budget control. Add tiderag for RAG. Both optional.

Create a Tool

import { defineTool } from '@carloscortezcloud/tinkuy-agent';

const myTool = defineTool({
  name: 'search',
  description: 'Search for information',
  parameters: { type: 'object', properties: { query: { type: 'string' } } },
  execute: async (args) => { /* your logic */ },
});

Create an Agent

import { Agent } from '@carloscortezcloud/tinkuy-agent';
import { StyrRouter } from '@carloscortezcloud/styrr-llm';

const agent = new Agent({
  router: new StyrRouter({ apiKey: process.env.OPENROUTER_API_KEY, models: [{ id: 'nvidia/nemotron-3-ultra-550b:free' }] }),
  tools: [myTool],
  systemPrompt: 'You are a helpful assistant.',
});

const result = await agent.run('Find info about serverless');
console.log(result.text);

2Architecture

User prompt
    |
    v
Agent (Tinkuy) ─── uses ──→ Router (Styrr)
    |                              |
    |─── optional ──→ Guard (Sayay) |
    |                              |
    v                              v
Tool Loop:                    Model Chain:
  LLM → tool_calls            model_1 → 429? → model_2 → 404? → model_3
  execute tools                (auto-fallback)
  feed results back
  repeat until text response
    |
    v
Final answer + metadata

3Packages

?FAQ

Do I need all 4 packages?

No. Use Tinkuy + Styrr as minimum (agent + routing). Sayay is optional (budget). TideRAG is optional (knowledge base). Each is independent.

Does it work with Strands Agents or LangChain?

Yes. Styrr can be used as a model provider for Strands or ChatModel for LangChain. Sayay can be a Strands Plugin. They complement, not compete.

Is it really free forever?

The packages are Apache 2.0 (free forever). Infrastructure cost depends on Cloudflare free tier limits (5M vectors, 100K req/day). For most projects this is plenty.

Can I use it without Cloudflare?

Styrr, Sayay, and Tinkuy work anywhere (Node, Deno, Bun, Lambda). Only TideRAG requires CF Workers (Vectorize + Workers AI + D1 are CF services).

How is this different from LangChain?

Tinkuy is 200 lines with 0 deps. LangChain is a full platform with many dependencies. Use Tinkuy when you want the smallest possible agent on edge. Use LangChain when you need a complete ecosystem with persistence, streaming, and multi-agent.