Daily recap for May 18, 2026

Today we published a new step-by-step tutorial for small-business AI: an Anthropic RAG pipeline that indexes SharePoint document libraries so teams can find answers without keyword-guessing. We also shipped two monorepo ideas that package common TypeScript patterns — a string utility library and a structured function-calling framework for AI agents — plus 23 new npm packages across three families.

New tutorials

Anthropic RAG Pipeline for SharePoint Knowledge Search

This recipe builds a hybrid retrieval-augmented generation pipeline that plugs into your SharePoint document libraries, combining BM25 keyword search and vector similarity to answer employee questions using Claude. It includes budget controls, query caching, and an escalation path to Slack when answers fall below a confidence threshold — a practical knowledge base that reduces time spent digging through scattered files.

Read the tutorial → | Download the code (zip)

Built with Anthropic, @reaatech/hybrid-rag, and Next.js · 119 tests, 100% coverage.

String Utils: Minimal TypeScript String Helpers

Two small, well-tested packages of string helpers — one with pure functions and one with a tiny CLI wrapper. Designed for TypeScript projects that need reliable string manipulation without pulling in heavy libraries.

Read the tutorial →

Structured Function Calling for AI Agents

A set of reusable middleware for enforcing schema contracts, validating tool outputs, and orchestrating function calls across multiple LLMs. Originally built with DeepSeek's DeepSeek-v4-Pro, these packages help you keep AI agent behavior predictable and testable.

Read the tutorial →

New repos

agent-runtime

A composable agent runtime for building and deploying autonomous agent systems. Provider-agnostic, with support for single- and multi-agent coordination, Zod schemas, and a model factory. Browse the code.

Building blocks shipped

multi-tenant-mcp

A full suite for building multi-tenant Model Context Protocol (MCP) servers — from tenant resolution and config isolation to rate limiting and cost accounting. The family covers the chain: @reaatech/multi-tenant-mcp-types provides shared interfaces, @reaatech/multi-tenant-mcp-tenant-resolver extracts identity from requests, @reaatech/multi-tenant-mcp-middleware enforces access control, and @reaatech/multi-tenant-mcp-cost-accounting tracks per-tenant usage. Browse the catalog.

otel-cost-exporter

Calculate and export real-time LLM cost metrics from OpenTelemetry spans. @reaatech/otel-cost-exporter-core defines the types and schemas, @reaatech/otel-cost-exporter-pricing bundles provider pricing data, @reaatech/otel-cost-exporter-calculator normalizes models and computes costs, and the main @reaatech/otel-cost-exporter package integrates as a SpanProcessor. A CLI is also available. Browse the catalog.

otel-genai-semconv

OpenTelemetry GenAI semantic convention instrumentation for all major LLM providers. @reaatech/otel-genai-semconv-core provides the types and span builders, while provider-specific packages instrument the Anthropic, OpenAI, Bedrock, and Vertex AI SDKs. Utilities for token estimation, PII redaction, and exporters for Arize Phoenix, Langfuse, and Google Cloud Trace round out the set. Browse the catalog.

Everything we build is yours to keep — browse the full catalog at reaatech.com/products.