How midday-ai/ai-sdk-tools works
Project type and purpose
• This is a Git-managed monorepo containing an open-source SDK (“AI SDK Tools”) and two Next.js applications (an example app and a documentation website).
• It exposes modular packages (agents, cache, artifacts, memory, store, devtools, debug, ai-sdk-tools, etc.) for building production-ready AI workflows, along with sample integrations.
File structure
• Top-level directories:
• .github/ – CI/CD workflows for beta and release.
• apps/ – Next.js “example” (AI chat sandbox) and “website” (docs & demos).
• packages/ – NPM-bundled libraries (agents, cache, artifacts, memory, store, devtools, debug, ai-sdk-tools).
• scripts/ – Publish, release, pre-publish, version bump, etc.
• .changeset/ – Versioning & changelog management.
• Each package has its own tsconfig.json and tsup.config.ts for building.
• Apps share a common Next.js structure (pages or app directory), import the SDK packages, and include API routes (chat, transcription, metadata).
README for additional insights
• Describes a “unified package” plus individual installs for each sub-package.
• Highlights key features: state management, debugging tools, structured streaming (artifacts), intelligent agents, caching, persistent memory.
• Directs users to the website for interactive demos and docs.
Elements
• From root → CI/CD → published packages.
• From Example App API routes → LLM providers, cache, memory.
• From Devtools → UI panel.
• Green for published packages,
• Blue for apps,
• Gray for scripts/CICD,
• Orange for external services.
Detailed capture of essential elements
• Data flow: user chat UI → chat hook (store) → agent orchestrator → tools (cache, memory, artifacts) → LLM → back through store → UI.
• Devtools interceptor tapped into that same data flow for debugging.