fly.toml

# fly.toml app configuration file for deepgram-python-stt
# Note: SocketIO requires sticky sessions for multi-machine deployments.
# With min_machines_running = 1 and auto_stop_machines = false, a single machine
# handles all connections, avoiding the need for Redis-backed session storage.
# If you scale beyond 1 machine, add sticky sessions via fly.io HTTP service options.

app = 'deepgram-python-stt'
primary_region = 'iad'

[build]

[env]
  PORT = "8080"

[http_service]
  internal_port = 8080
  # force_https required so browsers allow microphone access (getUserMedia needs HTTPS)
  force_https = true
  # Keep alive to avoid cold starts during demos
  auto_stop_machines = false
  auto_start_machines = true
  min_machines_running = 1

  [http_service.concurrency]
    type = "connections"
    hard_limit = 50
    soft_limit = 25

  [[http_service.checks]]
    grace_period = "10s"
    interval = "15s"
    method = "GET"
    path = "/"
    timeout = "5s"

[[vm]]
  memory = '512mb'
  cpu_kind = 'shared'
  cpus = 1