← Yellemo시행 2026-09-02

Google API Limited Use

The use of raw or derived user data received from Workspace APIs will adhere to the Google User Data Policy, including the Limited Use requirements.

Yellemo does not use Google user data (raw, aggregated, or derived) received from Workspace APIs or Photos APIs to create, train, or improve foundational or generalized machine learning or artificial intelligence models. Yellemo does not transfer that data to third-party AI/ML services. Yellemo does not sell Google user data.

회사는 Google Workspace API 또는 Photos API로 받은 사용자 데이터(원본·집계·파생)를 일반 AI/ML 모델의 생성·학습·개선에 쓰지 않으며, 제3자 AI 서비스로 이전하거나 판매하지 않습니다. 상세 개인정보 처리는 개인정보처리방침 §5를 따릅니다.

Workspace and Photos APIs we use

  • Google Calendar API only (optional): calendar.events.readonly plus email so the signed-in user can overlay their own primary calendar events in Yellemo. Read-only. We do not create, update, or delete events on Google Calendar.
  • Google Photos API: not used. We do not request Photos scopes.
  • Sign-in: OpenID openid email profile for account linking only. That is not used as AI training data.

Data isolation (no transfer of Workspace data to AI)

  • Calendar API payloads are stored as per-user overlay events (source=google) and shown only to the connecting user. Workspace members do not see another person's Google overlay.
  • AI features (memo organize, case brief, inquiry draft, "일정으로" calendar draft from a notebook block) send only text the user selected in the Yellemo notebook. They do not attach Google Calendar API rows, tokens, or Photos data.
  • The API rejects LLM prompts that look like exported Calendar API records (internal google:<userId>:<eventId> identifiers or source:google JSON).
  • Speech-to-text uses the Google Gemini / Interactions API with audio the user recorded in Yellemo. That is not Workspace Calendar or Photos user data.

Third-party AI providers and Yellemo plans

Text AI is called only through OpenRouter, Inc. (United States), an aggregator / model hub. Yellemo does not use DeepSeek. Yellemo does not call OpenAI, Anthropic, Upstage, or xAI APIs directly. Settings fields for those vendors are placeholders except OpenRouter (live) and Google AI Studio (live for speech-to-text only).

User plans and which catalog they may select:

  • Guest, Free, Student, Basic: standard OpenRouter catalog — Gemini 3.5 Flash Lite (Google), Solar Pro 4 (Upstage), GPT-4o mini (OpenAI), Claude 3 Haiku (Anthropic), Grok Build 0.1 (xAI). Monthly included AI usage applies.
  • Pro, Team, Enterprise: the standard catalog plus Gemini 3.5 Flash, Gemini 3 Pro, GPT-4o, Claude Sonnet 4.6, Claude Opus 4.6, Grok 4.5.
  • BYOK (optional):the same catalogs, billed to the user's own OpenRouter key. Still no Workspace/Photos API data in prompts.

Multi-model hub, upstream models, training restrictions

  • Hub: OpenRouter https://openrouter.ai/api/v1/chat/completions
  • Upstream models (current catalog slugs): google/gemini-3.5-flash-lite, upstage/solar-pro4, openai/gpt-4o-mini, anthropic/claude-3-haiku, x-ai/grok-build-0.1, and Pro slugs google/gemini-3.5-flash, google/gemini-3-pro-preview, openai/gpt-4o, anthropic/claude-sonnet-4.6, anthropic/claude-opus-4.6, x-ai/grok-4.5
  • API payload flag: every OpenRouter chat completion includes provider.data_collection: "deny" so OpenRouter must not route the request to endpoints that collect prompts for generalized model training. If a selected model cannot be served under that constraint, Yellemo falls back to another catalog model.
  • Account / dashboard:Yellemo's OpenRouter account is used for included-plan traffic. Users who paste their own OpenRouter key (BYOK) are billed on their OpenRouter account; we still send data_collection: deny and still omit Google Workspace/Photos data.

DeepSeek

Yellemo does not use DeepSeek. It is not in the model picker, not called via OpenRouter or a direct API, and not self-hosted. A previously stored DeepSeek preference is mapped to the default catalog model (Solar Pro 4).

Self-hosted or offline models

Yellemo does not operate a self-hosted or offline third-party foundation model against Google user data. All text AI inference for production is via OpenRouter HTTPS. A local heuristic stub may run when no LLM key is configured; it does not transmit data to a model provider.

Contact

Privacy: privacy@yellemo.com