Baidu File Translate

Baidu trans-cli file/document translation for 文件翻译、文档翻译、PDF/Word/Excel/PPT/TXT/HTML 翻译 while preserving original formatting and layout. Powered by trans-cli (Baidu LLM File Translation API). Trigger when the user uploads or references a file and asks to translate it, or asks to convert a document's language.

baidu-file-translate — Agent Reference

File translation is async. Always use the submit → poll loop below. Never rely on --wait without a timeout — it may block indefinitely in an Agent framework.

Step 1: trans file submit <file> --from auto --to <lang> --json
        → exit ≠ 0 → ⚠ STOP if exit 2 (auth/config); handle others (see Error Handling)
        → exit 0   → record request_id from stdout

Step 2: every 5 s: trans file query <request_id> --json
        → exit ≠ 0              → handle error; DOC_FAILED = don't retry
        → exit 0, status=processing → keep polling (set Agent timeout ≤ 10 min)
        → exit 0, status=done       → ⚠ confirm with user before downloading
                                       then: trans file query <request_id> --download --output ./
        → WAIT_TIMEOUT              → ⚠ STOP — save request_id, report to user

Set an Agent-level timeout (e.g. 10 min). On WAIT_TIMEOUT, save the request_id and report to user — the job is still running and can be resumed later.

---

trans file submit <file>

trans file submit report.pdf --from zh --to en --json
trans file submit report.pdf --to en --reference "Use academic tone"

Flag	Default	Description
--from	auto	Source language (auto-detect)
--to	zh	Target language — auto is not accepted
--reference	—	Custom translation instruction, ≤ 1000 Unicode code points
--trans-image	false	Also translate text inside images
--wait	false	Block until done (always pair with --wait-timeout)
--wait-timeout	0 (no limit)	Max wait time for --wait mode

trans file query <requestId>

trans file query REQ123 --json
trans file query REQ123 --download --output ./output/

Flag	Default	Description
--download	false	Download the translated file when done
--output	./	Download directory (used with --download)
--wait	false	Block until done
--wait-timeout	0 (no limit)	Max wait time for --wait mode

---

JSON Contract

submit success (stdout):

{"request_id": "REQ123", "status": "submitted"}

query — in progress (stdout):

{"request_id": "REQ123", "status": "processing", "name": "report.pdf", "from": "zh", "to": "en"}

query — done (stdout):

{"request_id": "REQ123", "status": "done", "file_url": "https://...", "char_count": 1000, "amount": 70}

With --download, the done response gains:

{"local_path": "/absolute/path/to/report_en.pdf"}

error (stderr, exit ≠ 0):

{"code": "AUTH_FAILED", "message": "...", "help_url": "https://fanyi-api.baidu.com/manage/apiKey"}

help_url is omitted when not applicable.

---

Error Handling

exit ≠ 0
├── CONFIG_MISSING (exit 2) → guide user to set TRANS_API_KEY
├── AUTH_FAILED    (exit 2) → key invalid/expired; check fanyi-api.baidu.com/manage/apiKey
├── QUOTA_EXCEEDED (exit 3) → balance exhausted; recharge, then retry
├── RATE_LIMITED   (exit 3) → back off 30 s, then retry
├── NETWORK_ERROR  (exit 4) → check connectivity, retry
├── DOC_FAILED     (exit 3) → server-side failure; do NOT retry
├── INVALID_INPUT  (exit 1) → fix file path / format / --to value
└── WAIT_TIMEOUT   (exit 1) → job still running; save request_id and resume later

stderr carries errors only — there are no progress events on stderr.

---

Supported Formats

docx / doc / pdf / xlsx / xls / pptx / ppt / html / htm / txt / xml / md

file_url expires after 30 days.