Providers are external executables that GARM calls to create and manage runner instances in a target infrastructure. GARM delegates all instance lifecycle operations (create, delete, start, stop, list) to the provider binary.

The GARM Docker image includes pre-built binaries for all providers in /opt/garm/providers.d/.

Add a [[provider]] section to config.toml:

[[provider]]
  name = "lxd_local"
  provider_type = "external"
  description = "Local LXD installation"
  [provider.external]
    provider_executable = "/opt/garm/providers.d/garm-provider-lxd"
    config_file = "/etc/garm/garm-provider-lxd.toml"

You can configure multiple providers to offer different infrastructure options. Each pool is tied to one provider.

By default, GARM passes a clean environment to providers, consisting only of the variables defined by the provider interface. This is intentional -- providers should be self-contained and not depend on the host environment.

However, some providers need access to host environment variables for authentication. For example, the AWS provider may need AWS_* variables for IAM role-based authentication (EC2 instance profiles), and the Azure provider may need variables for managed identity. Use environment_variables to pass these through:

[provider.external]
  provider_executable = "/opt/garm/providers.d/garm-provider-aws"
  config_file = "/etc/garm/garm-provider-aws.toml"
  environment_variables = ["AWS_"]

This passes all environment variables starting with AWS_ to the provider. You can also specify exact variable names (e.g., ["AZURE_CLIENT_ID", "AZURE_TENANT_ID"]).

garm-cli provider list

You can create pools on different providers for the same repository. This is useful for:

• Cost optimization: Use cheap on-prem LXD for most jobs, overflow to cloud
• Capability matching: GPU workloads on specific providers, regular jobs on others
• Multi-cloud resilience: Spread across providers for availability

# On-prem pool with high priority
garm-cli pool add --repo <ID> --provider-name lxd_local --priority=10 --tags ubuntu ...

# Cloud overflow pool with lower priority
garm-cli pool add --repo <ID> --provider-name aws_ec2 --priority=1 --tags ubuntu ...

With pack balancing, GARM fills the high-priority LXD pool first and only creates cloud runners when LXD is full.

Providers are executables that respond to GARM commands passed via the GARM_COMMAND environment variable. For details on building your own provider, see the External Provider Interface documentation.