duhub-cli

Command-line client for DUhub: transcribe audio, synthesize speech, translate text, and diarize meetings — all billed to your API key.

du stt rec.wav -o transcript.json
du tts "Muraho neza nshuti yanjye." -o out.wav
du mt "Muraho" --src rw --tgt en
du diarize meeting.wav --format srt -o speakers.srt

Installation

Requires Python 3.11+.

pipx install duhub-cli
du --version

Alternatives:

pipx install git+https://github.com/Digital-Umuganda/duhub-cli.git
pip install git+https://github.com/Digital-Umuganda/duhub-cli.git   # plain pip
python -m du_cli.main --help                                     # without installing

Note: du shadows the Unix du (disk usage) on your PATH. If that bothers you, invoke it as python -m du_cli.main instead.

Quickstart

1. Create an account on the Dashboard and copy an API key from API Keys (shown once — treat it like a password).
2. Save it:

   du configure --api-key du_live_xxxxxxxxxxxxxxxx

3. Check connectivity:

   du health
   du health --detailed

4. Transcribe something:

   du stt rec.wav

Configuration

du configure stores settings in ~/.config/du/config.toml:

Source	Priority	Example
Environment	Highest (always wins)	DU_API_KEY, DU_BASE_URL
Config file	Middle	du configure --api-key … --base-url …
Built-in default	Lowest	base https://du-api.dev.umuganda.digital

du configure                                # show current state
du configure --api-key du_live_xxx          # set key
du configure --base-url http://localhost:8080  # point at local gateway
du configure --reset                        # delete the file
export DU_API_KEY=du_live_xxx DU_BASE_URL=https://du-api.dev.umuganda.digital

Prefer DU_API_KEY env var over --api-key — flags land in shell history.

Commands

Every command exits 0 on success, 1 on failure (2 on CLI usage errors), prints a summary line (units, cache, request-id), and accepts --json for a single machine-readable object on stdout ({"data", "units", "cache", "request_id"}). -o writes the raw payload to a file instead of stdout.

du health

Liveness (/health) or full readiness (--detailed, checks STT/TTS/MT backends — the one to try first when calls fail).

du stt — speech to text

Upload audio (WAV/FLAC/OGG, ≤ 30 MB), get {"text": "…"}. Billed per audio second.

du stt rec.wav -o transcript.json
du stt rec.wav --idempotency-key run-42 -o transcript.json

du tts — text to speech

Synthesize Kinyarwanda speech to a 24 kHz mono WAV (-o is required — audio never goes to stdout). Billed per word. --text-file overrides TEXT when both are given.

du tts "Muraho neza nshuti yanjye." -o out.wav

du mt — translation

Translate between Kinyarwanda (rw) and English (en). Billed per input word.

du mt "Muraho" --src rw --tgt en

du diarize — who spoke when

Upload audio, get speaker-labelled segments. --format renders the same result as json (default), txt, or srt. Billed per audio second.

du diarize meeting.wav --format srt -o speakers.srt

The gateway /diarize route rolls out separately from the CLI. If it is not deployed yet, the CLI says so plainly instead of dumping a 404.

Scripting

--json prints exactly one JSON object — pipe-safe:

TEXT=$(du stt rec.wav --json | jq -r .data.text)

for f in clips/*.wav; do
  du stt "$f" -o "${f%.wav}.json" || echo "FAILED: $f" >&2
done

Errors

The CLI never dumps tracebacks — every failure is error + hint:

You see	Cause	Fix
invalid key (401)	Missing/unknown/revoked key	Copy a fresh key from dashboard API Keys
insufficient balance (402)	Wallet empty	Top-up via dashboard Wallet
organization pending review (403)	Org not approved yet	Contact your org admin
file exceeds 30 MB	Upload guard	Compress or split the audio
server busy (saturated) (503)	GPU at concurrency cap	Wait Retry-After seconds, retry
backend unreachable (502)	Model container down	du health --detailed to see which
cannot reach server	Wrong --base-url, offline, timeout	Check URL and connectivity

Development

uv sync        # or: python -m venv .venv && .venv/bin/pip install -e .
uv run pytest
uv run du --help