AssemblyAI CLI (assembly)

assembly runs AssemblyAI from the terminal: transcription, real-time streaming, voice agents, the LLM Gateway, history, and account management.

assembly <command> --help is the source of truth for flags. This skill covers the command map and the non-obvious operational rules; check --help before guessing a flag.

Critical: auth & environment

Authentication. A command needs a key resolved in this order:

1. ASSEMBLYAI_API_KEY environment variable
2. The OS keyring (populated by assembly login)

Get authenticated with either assembly login (browser sign-in; stores a key in the keyring) or by exporting ASSEMBLYAI_API_KEY. Run commands deliberately have no --api-key flag — that is on purpose, so keys never land in ps output or shell history. Do not look for one.

Environment binding. The backend environment is selected by --env (or AAI_ENV, or the profile's stored env). --sandbox is shorthand for --env sandbox000. The default environment is production. A credential is only valid against the environment that minted it — a sandbox key fails against production and vice-versa. If a freshly-working key suddenly returns auth errors, check you are on the same --env you logged in under.

Profiles. --profile <name> selects a named credential set. Global flags (--profile, --env, --sandbox, -v/--verbose) go before the subcommand: assembly --sandbox transcribe call.mp3.

Diagnostics. assembly -v <command> logs request-level diagnostics to stderr (HTTP requests and statuses); -vv adds wire-level detail (WebSocket frames, connection events). Secrets (API key, session JWT) are redacted from that output. Use this to debug auth/connectivity instead of guessing.

Output contract (read this before parsing output)

• Data goes to stdout; errors and progress go to stderr. Piping stdout is always safe.
• --json is auto-enabled when output is piped or the CLI detects an agent run, so you usually get machine-readable JSON on stdout for free. Pass --json explicitly to force it. Many commands also accept -o/--output to print a single field (e.g. -o text).
• Expected failures print a clean message to stderr and exit non-zero — never a traceback. Exit code reflects the error type.

Quick start

assembly login                      # browser sign-in (or: export ASSEMBLYAI_API_KEY=...)
assembly doctor                     # verify the environment is ready
assembly transcribe call.mp3        # transcribe a file
assembly transcribe call.mp3 -o text   # just the text, pipeline-friendly
assembly stream                     # live transcription from the mic
assembly init                       # scaffold a starter app

Building an app vs running a command

If the task is to build/create an app or project (a transcription app, live captions, or a voice agent app), that is assembly init — a scaffolder that writes a full starter project (pick the voice-agent template for an agent app). The verbs assembly transcribe, assembly stream, and assembly agent are run commands: they perform a one-off action in the terminal (e.g. assembly agent holds a live mic conversation) and produce no code. When someone says "build an agent," reach for assembly init voice-agent, not assembly agent.

Decision tree

• Build/scaffold an app (transcription, live captions, or a voice agent app) → assembly init — see references/setup.md
• Transcribe a file/URL/YouTube/podcast page, stream live audio, run a live voice agent, or query the LLM Gateway → references/transcription.md
• Browse past transcripts or streaming sessions → references/history.md
• Sign in/out, identity, balance, usage, rate limits, API keys, audit log → references/account.md
• Scaffold a starter app (init), diagnose setup (doctor), set up your coding agent's MCP + skills (setup) → references/setup.md

Anti-patterns

• Passing --api-key to a run command. It does not exist. Use assembly login or ASSEMBLYAI_API_KEY.
• Mixing a credential with the wrong --env. A sandbox000 key won't work against production. Log in and run under the same environment.
• Running before authenticating. No key → auth failure. Run assembly doctor to see exactly what's missing.
• Assuming pip install assemblyai-cli works. That PyPI name is squatted by an unrelated third party. Use the project's official install path, not that name.
• Parsing human output. Pipe stdout (auto-JSON) or pass --json / -o text rather than scraping the pretty-printed tables.
• Forgetting --show-code. transcribe, stream, and agent accept --show-code to print a ready-to-run Python SDK script for exactly the flags you passed — no API call made. Great for "how would I do this in code?".