Skip to content

v0.2.0 — M11 Multi-backend LLM Architecture

Choose a tag to compare

View all tags
@v0o0v v0o0v released this 20 May 13:00
· 28 commits to main since this release
v0.2.0
f68ef88
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

M11 (Multi-backend LLM Architecture) + backend cards + 5 fix. semver minor (additive only, breaking 0).

주요 변경

Multi-backend LLM (Phase 0~7)

6 backend wrapper — Ollama + Gemini + Claude + OpenAI + OpenRouter + HuggingFace + modality 별 chain (chat_image / chat_audio / text_embed) + 자동 fallback (transient만, hard 즉시 raise).

  • /settings 페이지 backend 카드 (enabled / API key / 모델 / 연결 테스트) + chain ▲/▼ 우선순위 + "체인 저장"
  • per-asset backend_image/audio/embed 컬럼 + MCP find_asset.backend_used + 검색 카드 🤖 배지
  • 6 default factory 자동 등록 — cfg.backends.<name>.enabled=True 만으로 사용. env fallback (GEMINI_API_KEY / ANTHROPIC_API_KEY / OPENAI_API_KEY / OPENROUTER_API_KEY / HF_TOKEN)

Backend 카드 가격·셋업 안내

각 backend 카드의 <details> block — 가격/셋업 안내 (ko/en) + API key 발급 페이지 direct link.

Gemini default 갱신

gemini-2.5-flash → gemini-3.1-flash-lite (output -40% 비용, 속도 +64%, intelligence +62%).

신규 의존성 4건 (runtime)

  • google-genai>=0.1
  • anthropic>=0.40
  • openai>=1.50
  • huggingface_hub>=0.24

설치

```bash
pipx install --upgrade assetcache-mcp

또는

uv tool install --upgrade assetcache-mcp
```

Verification

  • pytest -q → 1252 passed + 1 skipped + 53 deselected (v0.1.2 baseline 1079 + 173 신규)
  • pytest -m llm_integration 옵트인 (13 케이스 / 5 backend)

알려진 한계 (v0.2.x candidate)

  • AnalysisQueue → mark_asset_backends write hook (schema 만 준비)
  • registry 재시작 필요 (backend 설정 변경 즉시 반영 X)
  • API key 평문 저장 (OS keyring 없음)
  • Gemini Batch API (50% 비용, 24h SLO) — 다음 milestone

Full PR #16 · PyPI v0.2.0

Assets 2
Loading