Skynet Environment Variables

Skynet is configurable via environment variables. Some are shared by all modules, while others are specific to each one of them. You can skip setting the env variables for the disabled modules.

Shared Environment Variables

Name	Description	Default	Available values
ENABLE_METRICS	If the Prometheus metrics endpoint should be enabled or not	true	true, false
ENABLED_MODULES	Which modules should be enabled, separated by commas	summaries:dispatcher,summaries:executor,assistant	summaries:dispatcher, summaries:executor, assistant, streaming_whisper
BYPASS_AUTHORIZATION	If signed JWT authorization should be enabled	false	true, false
ASAP_PUB_KEYS_REPO_URL	Public key repository URL	NULL	N/A
ASAP_PUB_KEYS_FOLDER	Public key repository root path	NULL	N/A
ASAP_PUB_KEYS_AUDS	Allowed JWT audiences, separated by commas	NULL	N/A
ASAP_PUB_KEYS_MAX_CACHE_SIZE	Public key maximum cache size in bytes	512	N/A
LOG_LEVEL	Log level	DEBUG	DEBUG, INFO, WARNING, ERROR, CRITICAL
DISABLE_LLM_HEALTH_CHECK	Disables LLM health check if only OpenAI/Azure will be used	false	true, false

Assistant Module Environment Variables

Name	Description	Default	Available values
EMBEDDINGS_CHUNK_SIZE	Max size (in characters) of documents for which embeddings will be computed. Should be less than the max tokens of the embedding model	5000	N/A
EMBEDDINGS_MODEL_PATH	The path where the embeddings model is located.	BAAI/bge-m3	N/A
VECTOR_STORE_PATH	The default path where the vector store is saved locally	_vector_store_	N/A

Summaries Module Environment Variables

Name	Description	Default	Available values
ENABLE_BATCHING	Enable submitting jobs for inference while others are running. The actual batching needs to be supported by the underlying inference processor	true	true,false
LLAMA_PATH	The path where the llama model is located.	llama3.1	N/A
LLAMA_N_CTX	The context size of the llama model	128000	N/A
JOB_TIMEOUT	Timeout in seconds after which an inference job will be considered stuck and the app killed.	300	N/A
SUMMARY_MINIMUM_PAYLOAD_LENGTH	The minimum payload length allowed for summarization.	100	N/A
SKYNET_LISTEN_IP	Default ip address on which the webserver is started.	0.0.0.0	N/A
SKYNET_PORT	Default port on which the webserver is started.	8000	N/A

AI Studio Usage Reporting

When enabled, each completed inference job reports the tokens it consumed (as reported by the engine) to AI Studio's usage registry, so summary spend is counted alongside AI Studio's own in 8x8's customer-token metric. Disabled unless AI_STUDIO_USAGE_ENABLED is set; the URL and token must also be provided.

Name	Description	Default	Available values
AI_STUDIO_USAGE_ENABLED	Enable reporting per-job token usage to AI Studio.	false	true, false
AI_STUDIO_USAGE_URL	Full URL of the AI Studio usage endpoint (e.g. https://8x8.studio/api/usage).	NULL	N/A
AI_STUDIO_USAGE_TOKEN	Global service token (Bearer) authorising cross-tenant usage writes.	NULL	N/A
AI_STUDIO_USAGE_CHANNEL	Channel label recorded on each usage event.	skynet_summary	N/A
AI_STUDIO_USAGE_TIMEOUT	Timeout in seconds for the usage-reporting request.	5	N/A

Redis vars

Name	Description	Default	Available values
REDIS_EXP_SECONDS	After how many seconds will a completed job expire/be deleted from Redis	1800	N/A
REDIS_HOST	Redis host	localhost	N/A
REDIS_PORT	Redis port	6379	N/A
REDIS_USE_TLS	Use TLS when connecting to Redis	false	N/A
REDIS_DB_NO	Redis database number	0	N/A
REDIS_USR	Redis user if using user/pass auth	NULL	N/A
REDIS_PWD	Redis pass if using user/pass auth	NULL	N/A
REDIS_AWS_SECRET_ID	The ID of the secret to retrieve from AWS Secrets Manager	NULL	N/A
REDIS_USE_SECRETS_MANAGER	Use AWS Secrets Manager to retrieve credentials	false	N/A
REDIS_NAMESPACE	Prefix for each Redis key	skynet	N/A
REDIS_AWS_REGION	The AWS region. Needed when using AWS Secrets Manager to retrieve credentials.	us-west-2	N/A

OCI vars

Name	Description	Default	Available values
OCI_MAX_TOKENS	OCI Maximum output tokens (check https://docs.oracle.com/en-us/iaas/Content/generative-ai/pretrained-models.htm)	4000	N/A
OCI_MODEL_ID	OCI Model id	NULL	N/A
OCI_SERVICE_ENDPOINT	OCI Service endpoint	https://inference.generativeai.us-chicago-1.oci.oraclecloud.com	N/A
OCI_COMPARTMENT_ID	OCI Compartment ID	NULL	N/A
OCI_AUTH_TYPE	OCI Authorization type	API KEY	N/A
OCI_CONFIG_PROFILE	OCI Config profile	DEFAULT	N/A

S3 vars (used for RAG vector store replication)

Name	Description	Default	Available values
SKYNET_S3_ACCESS_KEY	S3 access key	NULL	N/A
SKYNET_S3_BUCKET	S3 bucket	NULL	N/A
SKYNET_S3_ENDPOINT	S3 endpoint	NULL	N/A
SKYNET_S3_REGION	S3 region	NULL	N/A
SKYNET_S3_SECRET_KEY	S3 secret key	NULL	N/A

Streaming Whisper Module Environment Variables

Name	Description	Default	Available values
BEAM_SIZE	Whisper beam size	1	N/A
WHISPER_MODEL_NAME	The Faster Whisper model name to use if you want to download it automatically at start-up. Don't define it if you intend to mount the model as a volume.	NULL	tiny, tiny.en, small, small.en, base, base.en, medium, medium.en, large-v2, large-v1. NOTE: check https://huggingface.co/SYSTRAN for model updates.
WHISPER_COMPUTE_TYPE	Quantization https://opennmt.net/CTranslate2/quantization.html	int8	int8, int8_float32, int8_float16, int8_bfloat16, int16, float16, bfloat16, float32
WHISPER_GPU_INDICES	Use multiple GPUs if available by specifying their indices separated by commas, e.g. 0,1 for two GPUs	0	N/A
WHISPER_DEVICE	Which device to use for inference. The default auto will automatically detect if a GPU is present and fall back to cpu if not.	auto	auto, cpu, gpu
WHISPER_MODEL_PATH	The path to the model folder	f'{os.getcwd()}/models/streaming_whisper'	N/A
WHISPER_RETURN_TRANSCRIBED_AUDIO	If the transcribed audio should be returned in the response as a base64 string for each segment. Useful for debugging.	false	true, false