Live

kyle.pericak.com/apps/llm-client/

Static export hosted under the main blog bucket (gs://kyle.pericak.com/apps/llm-client).

What it is

A browser chat UI that talks to any OpenAI-compatible /v1/chat/completions endpoint — llama-server, OpenRouter, etc. Chats and settings live in localStorage; there is no server-side storage.

Source: apps/llm-client/.

Tech stack

• Next.js 16 (static export) + React 19
• Tailwind v4 + shadcn/ui + Radix primitives
• zustand with persist for state
• Vitest (unit) + Playwright (E2E, mocked endpoints)
• pnpm workspace

Development

cd apps/llm-client
bin/install.sh              # first time: pnpm install + playwright browsers
bin/start-dev.sh            # dev server on :3100
bin/start-dev.sh 3200       # custom port
bin/kill-dev.sh             # stop
bin/test.sh                 # unit + E2E

Requires a running llama-server (or compatible) — default http://127.0.0.1:8080, configurable in the UI via the "Connected to" pill.

Deploy

cd apps/llm-client
npm run build
bin/prod-deploy.sh

bin/prod-deploy.sh rsyncs out/ to gs://kyle.pericak.com/apps/llm-client and sets Cache-Control: no-cache,no-store,must-revalidate on changed files so Cloudflare picks up updates immediately.

Only Kyle deploys to prod.

Versioning

Semver in package.json, injected at build via NEXT_PUBLIC_APP_VERSION in next.config.ts, exposed through src/lib/version.ts. Bump patch for fixes, minor for features, major for breaking changes.