SentinelLM/render.yaml at main · mohi-devhub/SentinelLM · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
# Render Blueprint — deploys SentinelLM's full stack: the FastAPI proxy, the
# Next.js dashboard, a managed Postgres database, and a Redis-compatible Key
# Value store (cache + rate limiting + cross-replica WebSocket fanout).
#
# Deploy: in the Render Dashboard, New → Blueprint → connect this repo.
# Render reads this file from the repo root automatically. During Blueprint
# creation you'll be prompted for every `sync: false` value below — have your
# LLM provider API key and a value for SENTINEL_API_KEY ready.
#
# Two values further down (NEXT_PUBLIC_API_URL / NEXT_PUBLIC_WS_URL, and
# SENTINEL_CORS_ORIGINS) assume the service names below are available and
# therefore resolve to https://sentinellm-api.onrender.com and
# https://sentinellm-dashboard.onrender.com. If you rename a service, or
# either default name is already taken elsewhere on Render, update the other
# service's matching env var to the actual assigned URL after first deploy.

databases:
  - name: sentinellm-db
    databaseName: sentinellm
    plan: basic-256mb # smallest paid tier — check the Dashboard for current free-tier eligibility
    postgresMajorVersion: "15"
    region: oregon

services:
  # ── Redis-compatible cache / rate-limit / WebSocket fanout backend ────────
  - type: keyvalue
    name: sentinellm-redis
    plan: free
    region: oregon
    ipAllowList: [] # empty = only other services in this Render account can reach it
    maxmemoryPolicy: allkeys-lru # fine for us — cache entries carry TTLs, rate-limit
    # counters are short-lived, and pub/sub messages aren't persisted anyway
    persistenceMode: "off" # nothing we store here needs to survive a restart

  # ── API — FastAPI proxy + 7-evaluator chain ────────────────────────────────
  - type: web
    name: sentinellm-api
    runtime: docker
    dockerfilePath: ./docker/Dockerfile.api
    dockerContext: .
    healthCheckPath: /health
    region: oregon
    plan: standard # torch + transformers + detoxify + sentence-transformers need
    # real memory — the 512MB starter plan is not enough
    autoDeployTrigger: commit
    # Multiple replicas are safe: the WebSocket feed uses Redis pub/sub fanout
    # (sentinel/ws/broadcaster.py) specifically so every replica's dashboard
    # connections stay correct regardless of which replica scored a request.
    # No disk is attached here on purpose — Render won't scale a service with
    # a disk attached — so the tradeoff is that ML model weights re-download
    # on every cold start instead of persisting. Bake them into the Docker
    # image at build time instead if that cost matters more to you than
    # horizontal scaling.
    # numInstances: 2
    envVars:
      - key: DATABASE_URL
        fromDatabase:
          name: sentinellm-db
          property: connectionString
      - key: REDIS_URL
        fromService:
          type: keyvalue
          name: sentinellm-redis
          property: connectionString
      - key: SENTINEL_CONFIG_PATH
        value: config.yaml
      - key: SENTINEL_ENV
        value: production
      - key: SENTINEL_API_KEY
        sync: false # required in production — see README § Security settings
      - key: SENTINEL_CORS_ORIGINS
        value: https://sentinellm-dashboard.onrender.com
      # Set only the ONE matching llm_backend.provider in config.yaml —
      # leave the other two blank in the Dashboard.
      - key: OPENAI_API_KEY
        sync: false
      - key: ANTHROPIC_API_KEY
        sync: false
      - key: GEMINI_API_KEY
        sync: false
      # Observability — both optional; the app runs with a no-op tracer and
      # log-only alerting when these are left unset. Uncomment and fill in
      # via the Dashboard to point at your own OTLP backend / Slack webhook.
      # - key: OTEL_EXPORTER_OTLP_ENDPOINT
      #   sync: false
      # - key: SENTINEL_ALERT_WEBHOOK_URL
      #   sync: false

  # ── Dashboard — Next.js ─────────────────────────────────────────────────────
  - type: web
    name: sentinellm-dashboard
    runtime: docker
    dockerfilePath: ./docker/Dockerfile.dashboard
    dockerContext: ./dashboard
    region: oregon
    plan: starter
    autoDeployTrigger: commit
    envVars:
      # Next.js bakes NEXT_PUBLIC_* vars into the client bundle at build time
      # (see docker/Dockerfile.dashboard) — Render automatically passes a
      # Docker-runtime service's envVars into the build step as build args,
      # so these apply correctly with no extra wiring.
      - key: NEXT_PUBLIC_API_URL
        value: https://sentinellm-api.onrender.com
      - key: NEXT_PUBLIC_WS_URL
        value: wss://sentinellm-api.onrender.com