render.yaml

# ============================================================================
# DeepTutor — Render Blueprint (one-click "Deploy to Render")
# ============================================================================
# Deploys the all-in-one DeepTutor image (FastAPI backend + Next.js frontend
# run together under supervisord) as a single Docker web service backed by a
# persistent disk at /app/data.
#
# SECRETS ARE ENVIRONMENT-ONLY. Every provider API key is declared below with
# `sync: false`, so Render prompts you for it at deploy time and stores it in
# its encrypted env-var store. Keys are NEVER written to the persistent disk,
# NEVER entered in the Settings UI, and NEVER returned to the browser — the UI
# only shows a "set via environment ✓/✗" indicator per provider.
#
# Minimal run: set at least ONE LLM key (e.g. OPENAI_API_KEY) plus the matching
# embedding key (OPENAI_API_KEY doubles as the OpenAI embedding key). Leave the
# rest blank. Add a search key (e.g. TAVILY_API_KEY) to enable web search.
#
# Deploy-path notes (see README "Deploy to Render"):
#   * Port: the image serves the Next.js UI on 3782 and the API on 8001 (the UI
#     proxies /api and /ws to the backend in-process). Confirm the web service
#     is routing to port 3782 in the Render dashboard after the first deploy.
#   * This Dockerfile is multi-stage; Render builds the final stage, which is
#     the lean `production` image (the `deploy` stage inherits from it). No
#     Docker "Target Stage" override is needed.
#   * Auth (login) and PocketBase are optional and JSON-driven on disk (the
#     entrypoint reads data/user/settings/*.json); the password secrets below
#     are wired via env, but enabling those features needs a one-time edit of
#     the settings JSON on the mounted disk. See the README.
# ============================================================================

services:
  - type: web
    name: deeptutor
    runtime: docker
    dockerfilePath: ./Dockerfile
    dockerContext: .
    plan: standard # 2 GB RAM; RAG indexing/parsing is memory-hungry. Bump if OOM.
    healthCheckPath: /
    autoDeploy: false
    disk:
      # Non-secret persistent state only: knowledge bases, working dirs, runtime
      # settings JSON, per-user workspaces, logs. No secret is ever stored here.
      name: deeptutor-data
      mountPath: /app/data
      sizeGB: 10
    envVars:
      # --- Non-secret config -------------------------------------------------
      - key: TZ
        value: UTC

      # --- Demo mode (per-IP rate limits) -----------------------------------
      # OFF by default. Set DEMO_MODE=true only on a PUBLIC demo service where
      # anyone can chat against your key, to cap per-visitor spend. See the
      # README "Demo mode" section. The two limit overrides are optional
      # (defaults: 15/min, 200/hour per IP).
      # - key: DEMO_MODE
      #   value: "true"
      # - key: DEMO_RATE_LIMIT_PER_MIN
      #   value: "15"
      # - key: DEMO_RATE_LIMIT_PER_HOUR
      #   value: "200"

      # --- LLM providers (set the ones you use; blank = disabled) -----------
      # Names come from deeptutor/services/provider_registry.py (env_key +
      # env_extras). One key covers that provider across LLM/embedding/TTS/etc.
      - key: OPENAI_API_KEY # OpenAI (LLM + embeddings + TTS/STT + image gen)
        sync: false
      - key: ANTHROPIC_API_KEY # Anthropic Claude
        sync: false
      - key: GEMINI_API_KEY # Google Gemini
        sync: false
      - key: DEEPSEEK_API_KEY # DeepSeek
        sync: false
      - key: GROQ_API_KEY # Groq
        sync: false
      - key: MISTRAL_API_KEY # Mistral
        sync: false
      - key: MOONSHOT_API_KEY # Moonshot / Kimi
        sync: false
      - key: DASHSCOPE_API_KEY # Alibaba DashScope / Qwen
        sync: false
      - key: MINIMAX_API_KEY # MiniMax
        sync: false
      - key: ZAI_API_KEY # Zhipu AI (GLM) — primary
        sync: false
      - key: ZHIPUAI_API_KEY # Zhipu AI (GLM) — alias used by some SDKs
        sync: false
      - key: STEPFUN_API_KEY # StepFun
        sync: false
      - key: XIAOMIMIMO_API_KEY # Xiaomi MiMo
        sync: false
      - key: NVIDIA_NIM_API_KEY # NVIDIA NIM
        sync: false
      - key: QIANFAN_API_KEY # Baidu Qianfan
        sync: false
      - key: OPENROUTER_API_KEY # OpenRouter gateway
        sync: false
      - key: AZURE_OPENAI_API_KEY # Azure OpenAI
        sync: false

      # --- Embedding / rerank providers (non-LLM) ---------------------------
      - key: COHERE_API_KEY # Cohere embeddings / rerank
        sync: false
      - key: JINA_API_KEY # Jina embeddings / rerank
        sync: false

      # --- Web-search providers ---------------------------------------------
      - key: TAVILY_API_KEY # Tavily search
        sync: false
      - key: BRAVE_API_KEY # Brave search
        sync: false
      - key: EXA_API_KEY # Exa search
        sync: false
      - key: SERPER_API_KEY # Serper (Google) search
        sync: false
      - key: PERPLEXITY_API_KEY # Perplexity search
        sync: false

      # --- Document parsing / RAG services ----------------------------------
      - key: MINERU_API_TOKEN # MinerU cloud PDF parsing
        sync: false
      - key: PAGEINDEX_API_KEY # PageIndex retrieval pipeline
        sync: false
      - key: LIGHTRAG_API_KEY # LightRAG server API key (shared by all LightRAG KBs)
        sync: false

      # --- Auth / PocketBase (optional; also needs on-disk JSON to enable) --
      - key: AUTH_PASSWORD_HASH # bcrypt hash for the built-in login (env-only secret)
        sync: false
      - key: POCKETBASE_ADMIN_PASSWORD # PocketBase admin password (env-only secret)
        sync: false