Weekly recap, May 18, 2026 – May 24, 2026

This week we shipped 30 production-grade tutorials, each one a complete recipe you can download and wire into your own stack. They span voice agents, lead intake, cost control, code sandboxes, document pipelines, and reliability suites — all with real budgets, real failovers, and real code. No contact-us-for-a-quote, no demos. Pick one that matches your problem and try it this afternoon.

New tutorials

Google Gemini Voice Agent for Twilio Call Handling

Answers inbound calls with a Gemini‑powered voice agent that can book appointments or look up FAQs — no 24/7 receptionist needed. Callers speak, Deepgram transcribes, Gemini classifies intent, a confidence router decides what to do, and ElevenLabs talks back over a bidirectional WebSocket stream. All calls are traced through Langfuse.

Read the tutorial →  |  Download the code (zip)

Built with @reaatech/agent-mesh, @reaatech/agent-memory, @reaatech/confidence-router, @reaatech/llm-cache, on Google Gemini · 91 tests, 98.95% coverage.

Azure AI Multi-Agent Handoff for Linear Issue Triage

Automatically classifies and routes incoming Linear issues to specialized agents — debug, docs, scheduling, or generic — then hands off context with a structured protocol. Circuit breakers protect API calls, a budget guard enforces cost limits, and every step is traced through Langfuse. Complex cases escalate to humans, but most are triaged before an engineer opens the issue.

Read the tutorial →  |  Download the code (zip)

Built with @reaatech/agent-handoff-protocol, @reaatech/agent-handoff-routing, @reaatech/agent-handoff-compression, @reaatech/confidence-router, @reaatech/circuit-breaker-agents, @reaatech/agent-budget-middleware, on Azure AI · 124 tests, 99.65% coverage.

AWS Bedrock Multi‑Agent Handoff for Slack Support Triage

Routes Slack support requests to specialist agents — billing, technical, or account management — using AWS Bedrock LLMs and keyword‑based confidence scoring. The handoff layer transfers conversation state cleanly, so no customer repeats themselves. Works as a Slack Bolt app you can drop into any workspace.

Read the tutorial →  |  Download the code (zip)

Built with @reaatech/agent-mesh, @reaatech/agent-handoff, @reaatech/agent-handoff-routing, @reaatech/confidence-router, @reaatech/agent-memory, on AWS Bedrock · 72 tests, 96.66% coverage.

Google Gemini AI Spend Control for SMBs

Adds real‑time cost tracking and budget enforcement to any Next.js app using Google Gemini. Pre‑flight checks block requests that would breach your hard cap, and you can auto‑downgrade to cheaper models when budgets tighten. A live /api/spend dashboard surfaces per‑tenant cost data, and OpenTelemetry exports everything to your monitoring stack.

Read the tutorial →  |  Download the code (zip)

Built with @reaatech/llm-cost-telemetry, @reaatech/agent-budget-engine, @reaatech/llm-cost-telemetry-aggregation, @reaatech/agent-budget-middleware, @reaatech/llm-router-core, @reaatech/agent-budget-otel-bridge, on Google Gemini · 79 tests, 96.55% coverage.

OpenAI Lead Intake with Adaptive Routing

Captures leads from web forms, parses attached PDFs and DOCX files, and routes each lead by confidence score. High‑confidence entries get handed off to a webhook; low‑confidence ones flag for review. Budget controls cap LLM spend per lead, and HubSpot sync keeps your CRM up to date.

Read the tutorial →  |  Download the code (zip)

Built with @reaatech/confidence-router, @reaatech/confidence-router-classifiers, @reaatech/agent-handoff, @reaatech/agent-budget-engine, on OpenAI · 199 tests, 100.00% coverage.

AWS Bedrock Code Sandbox for SMB Financial Report Generation

Lets your team ask natural‑language questions about financial data and get back formatted PDF and XLSX reports, all generated and executed safely in an isolated E2B sandbox. Budget engine caps total LLM spend per report, caching avoids recomputing identical queries, and structured output repair fixes broken JSON.

Read the tutorial →  |  Download the code (zip)

Built with @reaatech/agent-budget-engine, @reaatech/llm-cache, @reaatech/confidence-router, @reaatech/circuit-breaker-agents, on AWS Bedrock · 85 tests, 98.52% coverage.

OpenAI Voice Agent for Twilio Customer Support Calls

Handles inbound calls with OpenAI’s language model, classifies intent, enforces per‑call budgets, and falls back to voicemail when services degrade. Works as a Twilio‑connected Express app with Deepgram for speech‑to‑text and Cartesia for text‑to‑speech.

Read the tutorial →  |  Download the code (zip)

Built with @reaatech/confidence-router-core, @reaatech/circuit-breaker-agents, @reaatech/agent-budget-engine, @reaatech/agent-budget-middleware, @reaatech/llm-cache, @reaatech/mcp-server-tools, on OpenAI · 89 tests, 98.45% coverage.

Perplexity Code Sandbox for SMB Data Reporting

Small business analysts type a natural‑language question (like “show me last month’s sales by region”) and get back generated SQL or Python executed in a sandbox, with results and cost breakdown. Circuit breakers kill runaway executions, budget engine enforces spend caps, and structured output repair fixes malformed code.

Read the tutorial →  |  Download the code (zip)

Built with @reaatech/circuit-breaker-agents, @reaatech/agent-budget-engine, @reaatech/confidence-router, @reaatech/llm-cache, on Perplexity · 128 tests, 94.81% coverage.

Cohere Reliability Suite for SMB Customer Support

Wraps Cohere‑based support agents with model‑agnostic failover (to Azure OpenAI), circuit breakers, daily budget caps, and output repair. Semantically similar queries are cached, and duplicate webhook deliveries are caught, all orchestrated via Trigger.dev with Langfuse observability.

Read the tutorial →  |  Download the code (zip)

Built with @reaatech/llm-router-engine, @reaatech/circuit-breaker-agents, @reaatech/agent-budget-engine, @reaatech/llm-cache, @reaatech/idempotency-middleware, on Cohere · 117 tests, 94.48% coverage.

OpenRouter MCP Gateway for SMB Multi‑Tool Agents

A multi‑tenant MCP gateway that exposes SaaS tools (CRM, email, calendar) behind a unified auth layer, with per‑tenant rate limits, tool allowlists, and audit logging. OpenRouter powers the LLM‑based tool routing, and a CLI lets operators manage tenants and tokens.

Read the tutorial →  |  Download the code (zip)

Built with @reaatech/mcp-gateway-core, @reaatech/mcp-gateway-auth, @reaatech/multi-tenant-mcp-middleware, @reaatech/multi-tenant-mcp-rate-limiter, @reaatech/mcp-server-observability, on OpenRouter · 209 tests, 96.25% coverage.

xAI Grok Voice Agent for After‑Hours Customer Support

An AI receptionist that answers after‑hours calls, qualifies leads, and routes urgent issues via Twilio SMS. xAI Grok drives the conversation, LiveKit handles real‑time media, and a budget engine caps per‑call costs. Common responses are cached to reduce spend.

Read the tutorial →  |  Download the code (zip)

Built with @reaatech/confidence-router, @reaatech/agent-handoff, @reaatech/agent-budget-engine, @reaatech/llm-cache, on xAI Grok · 99 tests, 100.00% coverage.

Azure AI Spend Control for Multi‑Model SMB Workflows

Wraps Azure AI requests with real‑time budget checks, auto‑downgrades to cheaper models when caps approach, and exports cost telemetry to Langfuse via OpenTelemetry. Two API routes and a sidecar health server give you spend summaries and control.

Read the tutorial →  |  Download the code (zip)

Built with @reaatech/agent-budget-engine, @reaatech/agent-budget-middleware, @reaatech/agent-budget-otel-bridge, @reaatech/agent-budget-spend-tracker, @reaatech/agent-budget-llm-router-plugin, on Azure AI · 79 tests, 98.15% coverage.

vLLM Agent Eval Harness for Fine‑Tuned Model Quality

A CLI tool that runs automated evaluations on fine‑tuned models hosted behind a local vLLM server. Uses LLM‑as‑judge scoring (with GPT‑4) against golden datasets, tracks per‑trajectory cost, enforces regression gates, and pushes results to Langfuse. Six subcommands cover the full eval workflow.

Read the tutorial →  |  Download the code (zip)

Built with @reaatech/agent-eval-harness-cli, @reaatech/agent-eval-harness-judge, @reaatech/agent-eval-harness-cost, @reaatech/agent-eval-harness-gate, on vLLM · 126 tests, 100.00% coverage.

Agnostic AI Cost Control for QuickBooks Online SMBs

Monitors and caps spending across any LLM provider, then syncs categorized costs into QuickBooks Online as operational expenses. A Next.js app with five API routes provides real‑time per‑project budgets, model downgrading, and daily journal entry sync.

Read the tutorial →  |  Download the code (zip)

Built with @reaatech/agent-budget-engine, @reaatech/agent-budget-spend-tracker, @reaatech/agent-budget-llm-router-plugin, @reaatech/otel-cost-exporter, @reaatech/agent-budget-otel-bridge, on agnostic · 93 tests, 97.43% coverage.

AWS Bedrock Security Guardrails for Stripe Payment Agents

An MCP gateway that inspects every Stripe tool call from an AI agent and evaluates it against PII redaction, prompt injection, topic boundaries, and policy checks (like “no refund above $500 without human approval”). Claude Sonnet on Bedrock does the policy evaluation. Circuit breakers and idempotency guards protect against duplicate or malicious calls.

Read the tutorial →  |  Download the code (zip)

Built with @reaatech/guardrail-chain-guardrails, @reaatech/mcp-gateway-auth, @reaatech/mcp-gateway-rate-limit, @reaatech/circuit-breaker-agents, @reaatech/idempotency-middleware, on AWS Bedrock · 95 tests, 96.78% coverage.

Azure AI Reliability Suite for SMB AI Operations

Proactive incident detection and self‑healing for AI agents. Circuit breakers isolate failing tools, idempotency middleware prevents duplicate side effects, and Trigger.dev durable workflows run recovery playbooks. All backed by Azure OpenAI and traced through Langfuse.

Read the tutorial →  |  Download the code (zip)

Built with @reaatech/agent-runbook-agent, @reaatech/circuit-breaker-agents, @reaatech/idempotency-middleware, @reaatech/agent-runbook-mcp, on Azure AI · 102 tests, 97.70% coverage.

Vertex AI Document Pipeline for DocuSign SMB Contract Review

Extracts and validates key clauses from incoming DocuSign contracts — renewal dates, liability caps, payment terms — using hybrid RAG with Vertex AI Gemini. A golden‑set benchmark continuously measures extraction accuracy, and low‑confidence results flag for human review.

Read the tutorial →  |  Download the code (zip)

Built with @reaatech/hybrid-rag, @reaatech/hybrid-rag-embedding, @reaatech/hybrid-rag-ingestion, @reaatech/agent-eval-harness-golden, on Vertex AI · 126 tests, 98.86% coverage.

Anthropic Agent Mesh for E‑commerce Fraud Detection

Three specialist agents — transaction anomaly, account takeover, and chargeback risk predictor — collaborate through the mesh to evaluate each order. A confidence gate approves, flags for review, or blocks outright, and a human‑review agent steps in when specialists disagree. Everything is logged to Supabase.

Read the tutorial →  |  Download the code (zip)

Built with @reaatech/agent-mesh, @reaatech/agent-mesh-registry, @reaatech/agent-mesh-router, @reaatech/agent-mesh-session, @reaatech/agent-mesh-confidence, @reaatech/agent-mesh-observability, on Anthropic · 93 tests, 100.00% coverage.

Azure AI Code Sandbox for SMB Analytics

Accepts natural‑language analytics queries, generates code via Azure OpenAI, and executes it in an E2B sandbox with per‑user spending limits. A Next.js dashboard shows results and metrics, and circuit breakers prevent cascading failures if a script runs wild.

Read the tutorial →  |  Download the code (zip)

Built with @reaatech/agent-budget-engine, @reaatech/circuit-breaker-agents, @reaatech/agent-budget-middleware, @reaatech/circuit-breaker-core, @reaatech/circuit-breaker-persistence, on Azure AI · 65 tests, 93.63% coverage.

OpenAI Lead Intake Agent for SMB Real Estate

A Next.js API that accepts form submissions and attached PDF/DOCX files, extracts structured lead data via OpenAI’s Responses API, classifies buyer/seller/renter intent, prevents duplicates, and writes everything to HubSpot. Drop‑in ready for any real estate site.

Read the tutorial →  |  Download the code (zip)

Built with @reaatech/confidence-router, @reaatech/idempotency-middleware, @reaatech/confidence-router-core, @reaatech/idempotency-middleware-express, on OpenAI · 102 tests, 97.26% coverage.

AWS Bedrock Knowledge Agent for AppFolio Tenant Inquiries

A self‑service chat agent that answers tenant questions about policies, maintenance, and lease terms using AppFolio data. Hybrid RAG with Qdrant retrieves relevant documents, Claude on Bedrock answers with source citations, and a budget engine tracks per‑tenant spend.

Read the tutorial →  |  Download the code (zip)

Built with @reaatech/hybrid-rag, @reaatech/hybrid-rag-embedding, @reaatech/hybrid-rag-evaluation, @reaatech/agent-budget-engine, @reaatech/agent-eval-harness-golden, on AWS Bedrock · 128 tests, 100.00% coverage.

Ollama Code Sandbox for SMB Financial Analysis

Self‑hosted, on‑premises code execution that keeps sensitive financial data local. A local Ollama LLM generates Python or SQL, a Daytona sandbox runs it, and session‑level budget limits plus circuit breakers keep costs predictable.

Read the tutorial →  |  Download the code (zip)

Built with @reaatech/agent-budget-engine, @reaatech/circuit-breaker-agents, @reaatech/llm-cache, on Ollama · 56 tests, 95.27% coverage.

OpenAI Invoice Extraction for Xero SMB Accounting

Parses PDF and image invoices, extracts structured data using GPT‑5.2 vision, validates against Zod schemas, repairs malformed outputs, and syncs high‑confidence entries directly into Xero. Low‑confidence items flag for review; everything is archived in S3.

Read the tutorial →  |  Download the code (zip)

Built with @reaatech/hybrid-rag, @reaatech/hybrid-rag-embedding, @reaatech/agent-budget-engine, @reaatech/confidence-router, @reaatech/agent-eval-harness-golden, on OpenAI · 123 tests, 100.00% coverage.

Cohere Voice Agent for Shopify Order Status Checks

Answers inbound calls, extracts order numbers via Cohere’s Command model, looks up status in Shopify, and speaks the result — all protected by circuit breakers, per‑call budgets, and Zendesk escalation for complex cases. Observability is wired into Langfuse.

Read the tutorial →  |  Download the code (zip)

Built with @reaatech/agent-mesh, @reaatech/agent-handoff, @reaatech/circuit-breaker-agents, @reaatech/agent-budget-engine, on Cohere · 111 tests, 100.00% coverage.

OpenAI Multi‑Agent Handoff for E‑commerce Order Handling

Three specialist agents — order status, modifications, and cancellations — work together under Mastra orchestration. An LLM cache returns instant answers for common queries, a budget engine caps per‑conversation cost, and an idempotency layer prevents double refunds.

Read the tutorial →  |  Download the code (zip)

Built with @reaatech/agent-handoff, @reaatech/agent-mesh, @reaatech/llm-cache, @reaatech/agent-budget-engine, @reaatech/idempotency-middleware, on OpenAI · 109 tests, 97.26% coverage.

Anthropic Document Pipeline for SMB Lease Abstraction

Extracts key lease terms — dates, rent, clauses — from PDFs and DOCX files using Claude, with a hybrid RAG clause library for context. A confidence router auto‑populates abstracts for high‑confidence results and flags low ones for attorney review. Budget engine caps spend per document.

Read the tutorial →  |  Download the code (zip)

Built with @reaatech/hybrid-rag, @reaatech/confidence-router, @reaatech/agent-budget-engine, @reaatech/circuit-breaker-agents, on Anthropic · 146 tests, 99.70% coverage.

Mistral AI Agent Reliability Suite for SMB Customer Support

Wraps every Mistral tool call in a circuit breaker and idempotency middleware, exposes a health endpoint with auto‑generated Kubernetes probes, and runs Trigger.dev workflows for incident creation when a breaker trips. Keeps customer support agents up even when downstream APIs fail.

Read the tutorial →  |  Download the code (zip)

Built with @reaatech/circuit-breaker-agents, @reaatech/agent-runbook-health-checks, @reaatech/agent-runbook-incident, @reaatech/idempotency-middleware, @reaatech/agent-runbook-runbook, on Mistral AI · 63 tests, 98.51% coverage.

Vertex AI Voice Agent for Twilio Field Service Scheduling

Real‑time voice agent that books field service appointments into Google Calendar. Uses LiveKit’s Google plugin on Vertex AI for the brain, Twilio for telephony, and an agent handoff layer to route callers between booking, rescheduling, and FAQ specialists. Idempotency middleware prevents double‑bookings.

Read the tutorial →  |  Download the code (zip)

Built with @reaatech/agent-handoff, @reaatech/agent-budget-engine, @reaatech/idempotency-middleware, on Vertex AI · 135 tests, 98.95% coverage.

Anthropic MCP Server for SMB ERP Tool Access

Exposes HubSpot CRM operations as type‑safe MCP tools that any MCP‑compatible client (like Claude) can discover and invoke. Per‑tenant auth, tool allowlists, and token‑bucket rate limiting are built in, along with tamper‑evident audit logs.

Read the tutorial →  |  Download the code (zip)

Built with @reaatech/mcp-server-core, @reaatech/mcp-gateway-auth, @reaatech/mcp-gateway-allowlist, @reaatech/mcp-gateway-rate-limit, @reaatech/mcp-gateway-audit, on Anthropic · 90 tests, 95.50% coverage.

Vertex AI Reliability Suite for SMB Agent Operations

Layers circuit breakers, idempotency middleware, and automated recovery runbooks over Vertex AI agents. An Inngest durable workflow orchestrates backoff, health‑check recovery, and escalation when a breaker trips. Structured output repair fixes malformed LLM responses.

Read the tutorial →  |  Download the code (zip)

Built with @reaatech/circuit-breaker-agents, @reaatech/idempotency-middleware, @reaatech/agent-runbook-incident, on Vertex AI · 71 tests, 98.38% coverage.

Browse all solutions →