[server]
host = "0.0.0.0"
port = 7860
shutdown_grace_period_ms = 30000

[inference]
models_dir = "/var/lib/aximo/models"
default_offline_engine = "parakeet"
default_realtime_engine = "parakeet"

[inference.engines.parakeet]
kind = "parakeet"
path = "parakeet-tdt-0.6b-v3-int8"

[limits]
max_short_audio_requests = 2
max_short_audio_bytes = 15000000
max_short_raw_pcm_bytes = 1920000
max_short_audio_duration_ms = 60000
max_short_decoded_samples = 5760000

max_realtime_sessions = 4
max_short_inferences = 1
max_realtime_inferences = 1

max_realtime_session_bytes = 1920000
max_realtime_session_duration_ms = 60000
realtime_partial_min_interval_ms = 500
realtime_partial_min_chunk_bytes = 16000
realtime_event_channel_capacity = 64

short_inference_timeout_ms = 120000
realtime_partial_timeout_ms = 10000
realtime_final_timeout_ms = 120000

runtime_degrade_after_consecutive_failures = 3
runtime_degraded_policy = "readiness_only"
runtime_degraded_recovery_cooldown_ms = 30000