Open-source · Driven by AI agents

Raw in.
Cooked out.

Give an AI agent a video URL or local file. OpenBBQ turns it into an editable bilingual subtitle draft and a burned video, with deterministic workflow checks and an explicit path to human review.

From source video to finished subtitles

Raw 生肉 Cooked 熟肉
  1. 01

    Fetch

    get the video
  2. 02

    Transcribe

    write down every word
  3. 03

    Segment

    cut it into lines
  4. 04

    Translate

    into your language
  5. 05

    Review

    check every line
  6. 06

    Export

    make subtitle files
  7. 07

    Burn

    into the picture

Quality gates

Keep the workflow structurally sound.

OpenBBQ enforces the mechanical invariants that make a long agent run safe. Quality findings stay visible without turning every draft into a mandatory review queue.

CLI reference

Listening check

openbbq asr check

Low-confidence words and suspicious repetitions are surfaced as evidence. The agent can correct an obvious occurrence while translating; professional review remains available on demand.

→ ready: true

Translation check

openbbq translate check

Missing output is caught before delivery. Display budgets and terminology consistency remain visible as quality guidance for the draft.

→ ready: true

Professional review

openbbq review

Open the local editor when the work needs human-reviewed quality. Human edits are authoritative and the automatic workflow does not overwrite them.

→ ready: true

Deterministic delivery

openbbq agent finish

Structure, bounded batches, artifact freshness, and one-time export and burn are verified before OpenBBQ reports the draft ready.

→ ready: true

Agent-native

Give the routine work to an agent.

Tell Claude Code or Codex what you want to make. OpenBBQ gives the agent the commands, structured results, and checkpoints it needs to run the job.

Structured results

Commands can return JSON, so agents read results without scraping terminal output.

Teach it once

openbbq skill install gives your assistant the workflow and command patterns it needs.

Resume later

Each stage writes its result to the workspace. Stop today and continue from the same files tomorrow.

Work in small batches

Long transcripts are read and updated in bounded batches, which keeps the context focused and the work easy to review.

Read the agent guide
agent - zsh
$ openbbq skill install --agent claude
~/.claude/skills/openbbq-subtitles/
$ openbbq --json translate check zh --workspace demo
{ "ready": true }

The fansub craft

The agent handles the routine. You make the final call.

See it before you serve it

openbbq review opens a local editor with the video, its sound waveform, and the subtitles side by side. Edits save automatically, with split, merge, and undo. Once professional review begins, export respects that review state.

openbbq review --workspace demo --to zh

Looks like a fansub made it

Ready-made bilingual subtitle styles for landscape and vertical video, tidy line breaks, and a term list that keeps names consistent across episodes.

openbbq export --workspace demo --to zh --ass-preset fansub
defaultfansubfansub-compactmobile

Fire up the grill.

Requires Python 3.12+, uv, and FFmpeg. The automatic result is an editable draft; review it before publishing.

$ uv tool install 'openbbq[whispercpp]'