diff --git a/litellm/proxy/example_config_yaml/agent_session_ec2_config.yaml b/litellm/proxy/example_config_yaml/agent_session_ec2_config.yaml new file mode 100644 index 00000000000..9a32feafde0 --- /dev/null +++ b/litellm/proxy/example_config_yaml/agent_session_ec2_config.yaml @@ -0,0 +1,38 @@ +# Example: agent_session VM provider with EC2 (BYOC AWS). +# +# Provisions one EC2 per agent session in the *team's* AWS account. +# Per-team overrides (creds, subnet, IAM profile, AMI, instance type) live in +# `LiteLLM_AgentVMConfig`, populated via Settings UI (Epic G). +# +# See: +# litellm/proxy/agent_session_endpoints/vm_providers/ec2.py +# infra/ami/README.md (how to build the AMI) + +model_list: + - model_name: gpt-4 + litellm_params: + model: openai/gpt-4 + api_key: os.environ/OPENAI_API_KEY + +agent_settings: + vm_provider: ec2 # or `noop` for environments without AWS + sweep_interval_seconds: 30 # bootstrap/heartbeat/max-session sweepers tick + + ec2: + # Defaults applied when a team's `LiteLLM_AgentVMConfig` row leaves the + # field NULL. Captured from the B0 spike. + default_region: us-west-2 + default_ami_id: ami-CHANGEME # baked by `infra/ami/litellm-agent-runtime.pkr.hcl` + default_instance_type: t3.large + use_spot: true + max_session_minutes: 120 + bootstrap_timeout_seconds: 180 + heartbeat_timeout_seconds: 120 + + warm_pool: + enabled: false # B2 (LIT-2890) implements the warm-pool path + size: 2 + max_idle_minutes: 30 + +general_settings: + master_key: os.environ/LITELLM_MASTER_KEY