Skip to content

feat: add CLIProxyAPI backend and retranslation improvements#16

Merged
Mapleeeeeeeeeee merged 15 commits into
mainfrom
feat/cli-proxy-api-backend
Jul 3, 2026
Merged

feat: add CLIProxyAPI backend and retranslation improvements#16
Mapleeeeeeeeeee merged 15 commits into
mainfrom
feat/cli-proxy-api-backend

Conversation

@Mapleeeeeeeeeee

Copy link
Copy Markdown
Owner

Summary

  • add optional Docker Compose CLIProxyAPI setup for OpenAI-compatible CLI backends such as Antigravity/agy
  • strengthen Whisper prompt and translation/retranslation context handling
  • allow partial retranslation to update corrected source text and translated text
  • preserve audio when concatenating generated intro with burned subtitle video
  • document optional CLIProxyAPI setup and ignore local tool caches in Prettier

Tests

  • uv run ruff check src/ tests/
  • uv run ruff format --check src/ tests/
  • uv run mypy src/
  • uv run vulture src/bilingualsub --min-confidence=80
  • uv run pytest tests/unit -m unit --cov=bilingualsub --cov-report=xml --cov-fail-under=60
  • uv run pytest tests/integration -m integration --cov=bilingualsub --cov-report=xml --cov-fail-under=30
  • uv run pytest tests/e2e -m e2e --cov-fail-under=0
  • prettier --check "**/*.{js,jsx,ts,tsx,json,yaml,md,html,css}" --ignore-path .prettierignore

Mapleeeeeeeeeee and others added 15 commits June 26, 2026 16:58
Allow CLIProxyAPI (or any OpenAI-compatible proxy) to serve as the
translation backend by setting OPENAI_BASE_URL. This enables using
CLI subscription tokens (Codex, Antigravity, Copilot) instead of
paying for LLM API calls directly.

- Add openai_base_url setting with trailing-slash validator
- Build OpenAIChat model when proxy is configured, bypass Agno registry
- Skip API key validation when proxy handles auth
- Add 6 unit tests covering proxy/non-proxy model construction

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Pass video title and glossary source terms as Whisper's prompt
parameter to improve proper noun recognition during transcription.
Previously, metadata only flowed to the translator — now the
transcriber also benefits from contextual hints.

- Add build_whisper_prompt() to construct prompt from title + terms
- Add prompt parameter to transcribe_audio / _transcribe_single
- Add extract_source_terms() to parse glossary formatted text
- Wire prompt construction into pipeline before transcription call
- Add 12 unit tests (8 transcriber + 4 glossary)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
@Mapleeeeeeeeeee Mapleeeeeeeeeee merged commit e83a248 into main Jul 3, 2026
7 of 8 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant