Gloo AI TypeScript Examples

TypeScript examples for the Gloo AI platform API — a monorepo with two workspace packages:

Package	Description
chatbot/	Next.js streaming chatbot using the Completions V2 API
scripts/	CLI scripts for auth, chat, ingestion, search, and item management

Prerequisites

• Node.js LTS (.nvmrc provided — run nvm use or fnm use)
• pnpm (corepack enable pnpm if needed)
• Gloo AI credentials — get a client ID and secret at https://studio.ai.gloo.com/build/keys

Setup

# 1. Install dependencies
pnpm install

# 2. Configure credentials for the CLI scripts
cp scripts/.env.example scripts/.env.local
# Edit scripts/.env.local with your GLOO_AI_CLIENT_ID and GLOO_AI_CLIENT_SECRET

# 3. Configure credentials for the chatbot
cp chatbot/.env.local.example chatbot/.env.local
# Edit chatbot/.env.local with your GLOO_AI_CLIENT_ID and GLOO_AI_CLIENT_SECRET

Chatbot

A Next.js app demonstrating streaming markdown rendering with the Gloo Completions V2 API. Uses Vercel AI SDK v6 with a custom OpenAI-compatible provider, react-markdown for streaming token rendering, and Tailwind CSS v4.

Supports all three Gloo routing modes (AI Core, AI Core Select, AI Select) plus tradition-aware responses.

# Development server at http://localhost:3000
pnpm dev

# Production build
pnpm --filter gloo-chatbot build

Deploy to Vercel

Set Root Directory to chatbot and add the environment variables GLOO_CLIENT_ID and GLOO_CLIENT_SECRET.

CLI Scripts

Standalone scripts for exploring the Gloo AI platform APIs.

pnpm glooai:chat            # Chat completion (Completions V1)
pnpm glooai:ingest          # Content ingestion
pnpm glooai:items           # List items
pnpm glooai:items:metadata  # Item metadata
pnpm glooai:search          # Semantic search
pnpm glooai:jwt             # Decode and inspect access token
pnpm glooai:sonnet-4-repro  # Side-by-side: V1 Sonnet 4 vs. V2 Sonnet 4.5 / Haiku 4.5
pnpm glooai:whats-new       # Validate the 2026-05-26 release notes (models, cache, errors)

What's New — 2026-05-26 release validators

scripts/src/whats-new-2026-05-26.ts programmatically validates the subset of the Week of May 26, 2026 changelog that can be checked automatically from a single OAuth client. Three checks, each backed by a pure, unit-tested classifier:

#	Release-note claim	How it's validated
1	11 new models + Gemini 3.1 Flash Lite → GA	Asserts each model is present in the authoritative /platform/v2/models registry by display name (not by guessing alias IDs); the GA check additionally asserts the entry is no longer flagged preview.
2	cache_tokens / cache_hit_rate usage metrics	Sends a completion and validates both new usage fields are present and well-formed (non-negative tokens; hit-rate in [0, 1]).
3	Provider validation → actionable 4xx, not generic 500	Sends a deliberately invalid model and classifies the response (ACTIONABLE_CLIENT_ERROR vs the old GENERIC_SERVER_ERROR).

pnpm glooai:whats-new

A model reported ABSENT is surfaced, not silently dropped — it may indicate a not-yet-entitled / region-routing gap for this client rather than an unlaunched model (confirm in the Studio Model Explorer). Claims that are not cheaply auto-validatable from here (Studio billing/usage UI, streaming reliability, guardrail tuning, GlooCode, Data Engine org caps, docs restructuring) are intentionally out of scope.

Sonnet 4 V1→V2 reduced repro

scripts/src/sonnet-4-repro.ts is a side-by-side repro that compares the legacy V1 Messages endpoint using the deprecated Anthropic Bedrock inference profile us.anthropic.claude-sonnet-4-20250514-v1:0 against V2 Completions using the currently-supported Gloo aliases. It is intended as a reduced test case for validating model routing and lifecycle behavior after Anthropic's 2026-04-14 Sonnet 4 deprecation announcement.

Observed failure signatures on V1 with the deprecated model ID include both HTTP 5xx with {"detail":"Error generating response."} and HTTP 200 with an empty choices[0].message.content. The verdict classifier handles both — the exact signature varies by tenant/region routing.

The script runs three calls back-to-back against the same JWT so you can diff them:

#	Endpoint	Model	Purpose
1	/ai/v1/chat/completions	us.anthropic.claude-sonnet-4-20250514-v1:0	Deprecated Sonnet 4 under test (expected FAIL)
2	/ai/v2/chat/completions	gloo-anthropic-claude-sonnet-4.5	Supported Sonnet 4.5 alias on V2
3	/ai/v2/chat/completions	gloo-anthropic-claude-haiku-4.5	Supported Haiku 4.5 alias — faster/cheaper drop-in

Anthropic lifecycle context (publicly announced 2026-04-14):

• claude-sonnet-4-20250514 is marked Deprecated with a retirement date of 2026-06-15.
• API users may see degraded availability starting 2026-05-14.
• claude-sonnet-4-5-20250929 and claude-haiku-4-5-20251001 remain Active with no retirement sooner than 2026-09-29 / 2026-10-15 respectively.

Any workload still pinned to the retiring Sonnet 4 model ID should plan an upgrade path before 2026-05-14.

Development

The repo uses Turborepo for task orchestration. Root scripts delegate to workspaces:

pnpm build           # Build all packages
pnpm lint            # Lint all packages
pnpm typecheck       # Type-check all packages
pnpm test            # Run all tests
pnpm test:coverage   # Vitest coverage (scripts only, 70% minimum)
pnpm format          # Prettier — format all files
pnpm format:check    # Prettier — check formatting

Project Structure

├── chatbot/                 # Next.js streaming chatbot
│   ├── app/                 # App Router pages + API route
│   ├── components/          # Chat UI, message renderer, settings
│   └── lib/                 # Gloo auth + provider wrappers
├── scripts/                 # CLI scripts + tests
│   ├── src/                 # Auth, chat, ingestion, search, items
│   └── tests/               # Vitest unit tests
├── turbo.json               # Turborepo task config
└── pnpm-workspace.yaml      # Workspace packages