LINE Translate Worker

LINE bot translation worker powered by OpenAI and Cloudflare Workers.

Requirements

• Node.js >=20.0.0
• pnpm 11.7.0 via Corepack
• Cloudflare account + Wrangler
• LINE Messaging API channel
• OpenAI API key

Setup

1. Enable pnpm and install dependencies:
   • corepack enable
   • corepack prepare pnpm@11.7.0 --activate
   • pnpm install
2. Configure secrets:
   • pnpm exec wrangler secret put LINE_CHANNEL_SECRET
   • pnpm exec wrangler secret put LINE_CHANNEL_ACCESS_TOKEN
   • pnpm exec wrangler secret put OPENAI_API_KEY
3. Optional secret for precise mention matching:
   • pnpm exec wrangler secret put LINE_BOT_USER_ID
4. Deploy once to create the required Durable Object binding for atomic idempotency and rate limiting:
   • pnpm deploy

Runtime Vars (wrangler.jsonc)

• OPENAI_MODEL (current default: gpt-4o-mini)
• OPENAI_FALLBACK_MODEL (current default: gpt-4.1-mini)
• TRANSLATION_MODE (auto | ja2zh | zh2ja)
• TRANSLATION_STYLE (business | casual | neutral | polite)
• TRIGGER_MODE (all | mention | direct)
• TRIGGER_MENTION
• GROUP_TRANSLATION_ENABLED (true | false, default false)
  • true: translate every text message in groups and rooms.
  • false: groups and rooms translate only when the bot is explicitly tagged (@翻譯 or a command tag).
• DEBUG_LOG (true | false)
• MAX_INPUT_CHARS
• MAX_OUTPUT_TOKENS
• OPENAI_TIMEOUT_MS
• MAX_WEBHOOK_BODY_BYTES (default: 65536)
• RATE_LIMIT_PER_MIN
• IDEMPOTENCY_TTL_SECONDS
• ERROR_REPLY_ENABLED

The Worker explicitly sends store: false to OpenAI. It does not persist message text; the Durable Object stores only short-lived event IDs and rate-limit counters.

Local Run

• pnpm dev

Validation

• pnpm run typecheck
• pnpm test -- --run
• CI runs both commands on pull requests and pushes to main.

Deploy

• pnpm deploy
• Continuous deployment is not configured yet. Add it separately after defining the Cloudflare environment and repository secrets.

Debug Runbook

1. Set DEBUG_LOG to true.
2. Deploy.
3. Check logs with pnpm exec wrangler tail or Cloudflare dashboard.
4. For bot profile debug:
   • GET /debug/bot-info?debug=1
5. Set DEBUG_LOG back to false after troubleshooting.

Failure Modes

• Invalid signature: wrong LINE_CHANNEL_SECRET.
• OpenAI insufficient_quota: billing/quota issue.
• No reply in group mention mode:
  • Mention metadata may be absent for plain text tags.
  • Use command tags (@TWJP, @JPTW, @ENJP, @JPEN) or configure LINE_BOT_USER_ID.

Command Tags

• @TWJP, @JPTW, @ENJP, @JPEN
• Optional style suffix: -N (neutral/plain), -P (polite), -B (business)
• Examples:
  • @TWJP 明天麻煩你確認一下 -> defaults to polite Japanese (です・ます)
  • @TWJP-N 明天麻煩你確認一下
  • @TWJP-B 請協助安排下週會議
  • @JPTW-P お手数ですが、ご確認をお願いいたします