Daily recap for June 3, 2026

Today we published 14 new step-by-step tutorials for small-business AI, plus 30 building blocks updated underneath them. If you’re running a veterinary practice, auto repair shop, restaurant, SaaS, or any business that needs AI without a dedicated engineering team, there’s a recipe here ready to drop into your stack.

New tutorials

After-Hours Pet Emergency Triage Agent

After-hours phone calls are the biggest source of burnout for small vet practices. This tutorial builds a 24/7 voice triage agent that listens to pet owners, assesses urgency, and routes only true emergencies to the vet. It uses speech-to-text, intent classification, confidence-gated routing, and an LLM to generate structured recommendations — all traced through observability.

Read the tutorial → · Download the code (zip)

Built with @reaatech/voice-agent-core, @reaatech/voice-agent-telephony, @reaatech/voice-agent-stt, @reaatech/agent-mesh-classifier, @reaatech/agent-mesh-confidence, and @reaatech/guardrail-chain-guardrails, on agnostic · 74 tests · 99.24% coverage.

Auto-repair call intake agent

Service advisors at independent auto-repair shops are constantly interrupted by price-estimate calls. This solution builds a voice agent that handles the initial triage — listening to the customer’s problem, identifying the repair type, providing a ballpark estimate, and escalating to a human when needed. It combines speech-to-text, LLM-based repair assessment, and text-to-speech with Twilio for telephony.

Read the tutorial → · Download the code (zip)

Built with @reaatech/voice-agent-core, @reaatech/voice-agent-stt, @reaatech/voice-agent-tts, @reaatech/voice-agent-telephony, @reaatech/agent-handoff, and @reaatech/agent-handoff-routing, on agnostic · 119 tests · 100.00% coverage.

Per-tenant LLM cost tracking and chargeback

B2B SaaS platforms often absorb LLM costs across all tenants, eroding margins. This tutorial builds a system that attributes every API call to the right tenant, computes real-time spend metrics, and exports cost data to your billing infrastructure. It wires six REAA packages into a Next.js dashboard and REST API.

Read the tutorial → · Download the code (zip)

Built with @reaatech/agent-budget-spend-tracker, @reaatech/agent-budget-otel-bridge, @reaatech/llm-cost-telemetry, @reaatech/llm-cost-telemetry-exporters, @reaatech/otel-cost-exporter, and @reaatech/multi-tenant-mcp-cost-accounting, on agnostic · 74 tests · 100.00% coverage.

Anthropic Knowledge Agent for Shopify Product Q&A

Small store owners field repetitive product questions from website visitors. This agent connects directly to a Shopify store, uses hybrid RAG (vector + keyword search) to find relevant products, and answers with Anthropic’s Claude. Multi-turn conversations and user preferences persist across visits.

Read the tutorial → · Download the code (zip)

Built with @reaatech/hybrid-rag, @reaatech/agent-memory, @reaatech/session-continuity, and @reaatech/session-continuity-storage-dynamodb, on Anthropic · 111 tests · 99.64% coverage.

xAI Grok Observability for SMB AI Workflow Monitoring

If you’re using xAI Grok in your application, this tutorial gives you a single-pane dashboard for token usage, latency, errors, and cost. It wraps every Grok call with OpenTelemetry spans and real-time cost telemetry, then exports everything to Langfuse.

Read the tutorial → · Download the code (zip)

Built with @reaatech/otel-genai-semconv-core, @reaatech/llm-cost-telemetry, @reaatech/otel-genai-semconv-instrumentation, and @reaatech/llm-cost-telemetry-providers, on xAI Grok · 59 tests · 96.64% coverage.

Automated Industry Newsletter Curator

A consulting firm spends hours each week manually curating a client newsletter from 20+ RSS feeds. This pipeline fetches articles, scores them by relevance, generates AI summaries, and compiles a polished draft — cutting the task to near zero. Built with the hybrid RAG ecosystem and Hono.

Read the tutorial → · Download the code (zip)

Built with @reaatech/hybrid-rag, @reaatech/hybrid-rag-pipeline, @reaatech/hybrid-rag-retrieval, @reaatech/hybrid-rag-ingestion, @reaatech/hybrid-rag-observability, and @reaatech/hybrid-rag-mcp-server, on agnostic · 92 tests · 97.79% coverage.

xAI Grok Code Sandbox for Small Business Financial Modeling

SMB owners can describe a financial model in plain English and have Grok generate and execute the code in an isolated sandbox. The recipe enforces strict safety policies with a tool-use firewall, tracks costs, and exports results as spreadsheets.

Read the tutorial → · Download the code (zip)

Built with @reaatech/tool-use-firewall-core, @reaatech/llm-cost-telemetry, @reaatech/session-continuity, @reaatech/structured-repair-core, and @reaatech/confidence-router, on xAI Grok · 97 tests · 97.74% coverage.

Mistral AI Document Pipeline for NetSuite Invoice Automation

Finance teams waste hours keying invoice PDFs into NetSuite. This pipeline ingests PDFs, extracts text with OCR, parses line items and vendor details using Mistral AI, repairs any malformed JSON, and posts draft vendor bills back to NetSuite — with deduplication and full observability.

Read the tutorial → · Download the code (zip)

Built with @reaatech/media-pipeline-mcp-doc-extraction, @reaatech/structured-repair-core, and @reaatech/agent-memory, on Mistral AI · 99 tests · 100.00% coverage.

Configurable PII redaction at ingress and egress

A health-tech startup building a clinical agent must keep PHI from reaching the LLM. This dual-pipeline redacts PII at ingress (before the prompt) and egress (before the response), switchable between HIPAA, GDPR, and PCI rules — all auditable and running offline with regex-based detection.

Read the tutorial → · Download the code (zip)

Built with @reaatech/guardrail-chain, @reaatech/guardrail-chain-guardrails, @reaatech/guardrail-chain-config, @reaatech/guardrail-chain-observability, @reaatech/mcp-gateway-validation, and @reaatech/mcp-gateway-audit, on agnostic · 125 tests · 98.59% coverage.

Review Response Agent for Independent Restaurants

Restaurant owners spend hours manually replying to reviews across Yelp, Google, and TripAdvisor. This agent generates on-brand, consistent replies using the A2A protocol, with caching, guardrails, and memory so repeats stay fast and tone stays true.

Read the tutorial → · Download the code (zip)

Built with @reaatech/a2a-reference-core, @reaatech/a2a-reference-client, @reaatech/agent-memory, @reaatech/llm-cache, @reaatech/guardrail-chain, and @reaatech/a2a-reference-observability, on agnostic · 108 tests · 95.37% coverage.

Automated Blog Content Pipeline with SEO Scoring

A SaaS company wants to turn topic briefs into published posts with SEO gates. This pipeline generates the draft, validates markdown, lints for style, scores SEO factors, and delivers a report — all behind a single API call. Built with the agents-markdown ecosystem.

Read the tutorial → · Download the code (zip)

Built with @reaatech/agents-markdown, @reaatech/agents-markdown-validator, @reaatech/agents-markdown-reporter, @reaatech/agents-markdown-linter, @reaatech/agents-markdown-mcp-server, and @reaatech/agents-markdown-observability, on agnostic · 89 tests · 100.00% coverage.

Slack/email triage to priority inbox

A support team drowns in Slack DMs and emails. This agent reads inbound messages, classifies them P0/P1/P2, and drafts replies for urgent items. It uses intent classification with confidence gating and budget-aware routing so you never miss a critical issue.

Read the tutorial → · Download the code (zip)

Built with @reaatech/agent-mesh-classifier, @reaatech/agent-mesh-confidence, @reaatech/agent-handoff-routing, @reaatech/agent-budget-llm-router-plugin, and @reaatech/agent-mesh-utils, on agnostic · 86 tests · 99.58% coverage.

Persistent customer memory across support sessions

Your support agent forgets everything between calls. This tutorial gives it durable per-customer memory — previous conversations, open tickets, preferences — all stored with a token budget so context windows stay clean. Uses session-continuity packages and Anthropic Claude.

Read the tutorial → · Download the code (zip)

Built with @reaatech/session-continuity, @reaatech/session-continuity-storage-redis, @reaatech/session-continuity-storage-dynamodb, and @reaatech/session-continuity-tokenizers, on Anthropic · 89 tests · 100.00% coverage.

Automated API key rotation for LLM providers

Security policies demand key rotation, but manual changes cause outages. This sidecar rotates OpenAI and Anthropic keys across AWS Secrets Manager or Vercel Environment Variables without dropping a request, with audit logging and a CLI tool.

Read the tutorial → · Download the code (zip)

Built with @reaatech/secret-rotation-core, @reaatech/secret-rotation-sidecar, @reaatech/secret-rotation-provider-aws, @reaatech/secret-rotation-provider-vercel, @reaatech/secret-rotation-observability, and @reaatech/secret-rotation-types, on agnostic · 125 tests · 97.50% coverage.

Browse all solutions →

Building blocks shipped

Agent budget controller

The agent-budget-controller family received three patch releases. @reaatech/agent-budget-engine fixes a crash when policies omit auto-downgrade or disable-tools options; @reaatech/agent-budget-pricing now supports optional cache read/write token pricing for more accurate cost computation; and @reaatech/agent-budget-types migrated zod schemas from v3 to v4 without breaking public shapes.

Browse the family →

Classifier evals

All seven packages in the classifier-evals family received a security patch that updates vitest and @vitest/coverage-v8 to address a CI audit vulnerability. No public API changes.

Browse the family →

MCP contract kit

The mcp-contract-kit ecosystem got six patch releases, primarily to patch a critical vitest vulnerability (GHSA-5xrq-8626-4rwp). The validators package also picked up a yaml security bump and a fast-uri override. The CLI tool got a commander dep bump from 12 to 14.

Browse the family →

MCP server starter TS

The MCP server starter kit shipped two updates: @reaatech/mcp-server-core upgraded zod from v3 to v4 for security, and @reaatech/mcp-server-observability bumped OpenTelemetry packages to 2.x lines, including a migration from the deprecated new Resource(...) constructor.

Browse the family →

Media pipeline MCP

Three media pipeline providers received minor updates with underlying SDK upgrades. @reaatech/media-pipeline-mcp-deepgram migrated from Deepgram SDK 3.x to 5.x; @reaatech/media-pipeline-mcp-google bumped Google Cloud AI Platform and Document AI to major new versions; and @reaatech/media-pipeline-mcp-replicate went from replicate SDK 0.29.x to 1.4.0.

Browse the family →

Webhook relay MCP

The webhook relay system launched its first npm packages: @reaatech/webhook-relay-core provides shared types and encryption helpers; @reaatech/webhook-relay-storage handles SQLite persistence; @reaatech/webhook-relay-tools exposes 15 MCP management tools; @reaatech/webhook-relay-webhooks provides ingestion logic for Stripe, GitHub, and other sources; and @reaatech/webhook-relay-mcp bundles everything into a single CLI binary.

Browse the family →

@reaatech/agent-auth-proxy-client

Re-exports error classes from the core package, enabling consumers to perform instanceof checks on thrown errors.

View package →

@reaatech/circuit-breaker-core

Fixes two public-API bugs: the failureStrategy option was silently ignored, and ErrorThresholdStrategy would incorrectly accept an options object as its threshold. Both now work correctly.

View package →

@reaatech/llm-router-engine

Fixes a corrupted package.json dependency line that was breaking installs.

View package →

Browse the full catalog at reaatech.com/products.