Daily recap for May 15, 2026

Today we shipped nine new step-by-step tutorials for small-business AI, from a QuickBooks invoice pipeline to voice agents for Calendly and Cal.com. Twenty-four new packages landed underneath them, expanding our LLM routing, MCP contract testing, and MCP gateway families.

New tutorials

OpenRouter Agent Mesh for SMB Workflow Automation

This tutorial shows you how to build an Express server that acts as a central coordinator for multiple AI specialists. Incoming tasks are classified, assigned a confidence score, and routed to the right agent—all while tracking costs and preventing duplicate work. Perfect for small businesses that need to automate complex workflows across departments without a dedicated engineering team.

Read the tutorial → · Download the code (zip)
Built with OpenRouter, 75 tests, 95.82% coverage.

Azure AI Document Pipeline for QuickBooks Online Invoice Processing

Skip manual data entry: this recipe connects uploaded invoices to your QuickBooks Online account. It uses Azure’s AI to pull out line items, totals, and vendor details, then validates and cleans the data before posting. If you handle more than a handful of invoices each month, this pipeline can reclaim hours of admin time.

Read the tutorial → · Download the code (zip)
Built with Azure AI, 76 tests, 100% coverage.

Anthropic Voice Agent for Calendly Appointment Booking

Give your callers a voice agent that can check your real-time availability and book meetings directly in Calendly. It understands natural language like “next Tuesday at 3pm” and hands off to a human if the cost limit is reached. A practical way to cut down on phone tag without missing appointments.

Read the tutorial → · Download the code (zip)
Built with Anthropic, 100 tests, 97.66% coverage.

Mistral AI Multi-Agent Handoff for Linear Support Triage

If your support team uses Linear, this tutorial builds a system that triages tickets with AI and hands them off between specialist agents without losing history. Context is compressed and passed along, so customers don’t repeat themselves. Built with Mistral’s models for classification and reasoning.

Read the tutorial → · Download the code (zip)
Built with Mistral, 237 tests, 99.60% coverage.

AWS Bedrock Cost Control for Serverless AI Applications

Keep AWS Bedrock costs predictable with a budget proxy that caps spending per scope. When a limit approaches, it automatically downgrades to cheaper models or stops requests. Includes a dashboard to see exactly which models are consuming your budget. Essential for any small business running serverless AI on AWS.

Read the tutorial → · Download the code (zip)
Built with AWS Bedrock, 170 tests, 98.19% coverage.

Anthropic Security Guardrails for SMB AI Chatbots

Protect your AI chatbot’s users by adding PII redaction, injection blocking, and content safety checks. This middleware sits in front of any Anthropic-powered chat endpoint and logs every decision. You stay in full control—no third-party gateway needed.

Read the tutorial → · Download the code (zip)
Built with Anthropic, 137 tests, 100% coverage.

Ollama RAG Knowledge Base for Retail Inventory

Run a completely local question-answering system on your product documents. Staff can ask things like “Do we have Widget X in stock?” and get answers sourced from your catalog PDFs. Uses Ollama for the language model—no cloud costs or data leaving your network.

Read the tutorial → · Download the code (zip)
Built with Ollama, 106 tests, 100% coverage.

vLLM Security Guardrails for SMB API Gateways

Put a security layer in front of your self-hosted vLLM instance. It scrubs PII, blocks injection attempts, and monitors content—all configurable from a dashboard. If you’re running your own LLM server, this adds the safety checks you need without complexity.

Read the tutorial → · Download the code (zip)
Built with vLLM, 186 tests, 98.00% coverage.

Vertex AI Voice Agent for Cal.com Appointment Scheduling

Similar to the Calendly voice agent, but using Cal.com and Google’s Vertex AI. Callers can book appointments by phone, and the agent handles availability checks, booking confirmations, and PII stripping. Ideal for businesses using the open-source Cal.com platform.

Read the tutorial → · Download the code (zip)
Built with Vertex AI, 260 tests, 98.09% coverage.

Building blocks shipped

LLM Router

A suite for routing LLM requests with fallback, strategy selection, cost tracking, and an MCP-compatible server interface. @reaatech/llm-router-engine orchestrates model selection, @reaatech/llm-router-fallback adds resilience, and @reaatech/llm-router-strategies picks models by cost or latency. Seven packages in total landed in this family.

Browse the LLM Router catalog →

MCP Contract Kit

Validate and test Model Context Protocol servers before you deploy them. @reaatech/mcp-contract-core provides TypeScript interfaces and Zod schemas, @reaatech/mcp-contract-client is a pluggable MCP client, and @reaatech/mcp-contract-validators runs compliance, security, and performance checks. Six packages shipped.

Browse the MCP Contract Kit catalog →

MCP Gateway

The full-stack API gateway for Model Context Protocol servers. @reaatech/mcp-gateway-gateway is the Express-based server, @reaatech/mcp-gateway-auth handles API keys, JWTs, and OAuth2, and @reaatech/mcp-gateway-rate-limit enforces per-tenant quotas. Ten packages total, including caching, fan-out, and observability.

Browse the MCP Gateway catalog →

MCP Catalog

@reaatech/mcp-catalog-shared provides shared types and utilities for the MCP catalog ecosystem.

View package details →

Browse the full catalog at reaatech.com/products.