subbake 0.4.2

LLM-powered subtitle translation CLI with batch memory and review passes.

subbake

subbake is an agent-first subtitle translation CLI.

It translates .srt, .vtt, and line-based .txt subtitles with LLMs, while preserving subtitle structure and keeping long-running translation work recoverable through cache, translation memory, glossary, and resume state.

The main entrypoint is the interactive agent — reference files with @ and use / commands:

sbake

翻译 @episode01.srt         translate a subtitle file
翻译 @Season01              translate a folder as a series
/model chatgpt              switch model profiles
/session                    switch between agent sessions

No config file? The agent creates your first profile on startup.

For detailed setup and examples, use the project Wiki.

Why SubBake

• Agent workflow: run sbake, reference @file/@folder, switch profiles with /model or sessions with /session, toggle plan mode with Shift+Tab.
• Subtitle-safe: preserves ids, timing, cue settings, and line counts.
• Series support: translate a whole season with shared glossary and translation memory.
• Runtime repair: malformed model output is logged and repaired automatically.
• Review pass: high-risk batches receive targeted consistency review.
• Persistence: cache, run state, glossary, and translation memory live under .subbake.
• Config profiles: subbake.toml supports multiple provider presets for quick switching.

Agent Boundaries

The agent handles translation, series, profile switching, failure diagnosis, subtitle editing, and project-local file operations (create, append, replace, rename, delete).

It does not expose low-level file tools as direct commands — describe the task naturally and the agent selects the tool. In plan mode (Shift+Tab), mutating actions wait for /approve before execution. The agent is sandboxed to the project root; .git, .venv, .subbake, and __pycache__ are blocked. Destructive operations create backups under .subbake/agent/backups.

Install

pip install subbake

Then configure a provider profile. See the configuration example and the Wiki.

CLI Modes

Interactive agent:

sbake
sbake resume

Classic single-file command:

sbake translate input.srt --profile chatgpt

Series command:

sbake series ./Season01 --profile chatgpt

Credential and cleanup utilities:

sbake check-key --profile chatgpt
sbake clean .

Documentation

The Wiki is the primary user guide:

• Project Wiki
• Configuration example
• GitHub repository

Command help is also available from the CLI:

sbake --help
sbake translate --help
sbake series --help
sbake check-key --help
sbake clean --help