Secrets

Workers often need credentials, for example a token for an external API. Managed deployments read them from the workspace Secrets Manager: you bind a workspace secret to an environment variable, and the worker reads the value from its environment at boot. Only secrets from the deployment's workspace can be bound.

Bind a workspace secret to your deployment:

# /// script
# dependencies = [
#     "mistralai>=3.0",
# ]
# ///
import os

from mistralai.client import Mistral

client = Mistral(api_key=os.environ["MISTRAL_API_KEY"])

deployment = client.workflows.deployments.update_deployment(
    name="<your-deployment-name>",
    spec={
        "backend_spec": {
            "secrets": [
                {
                    "env_var_name": "GITHUB_TOKEN",
                    "reference": "secret:workspace:github_token",
                }
            ],
        }
    },
)
print(deployment.name)
Warning

Changing a secret's value does not restart the worker. The worker keeps the value it read at boot. To pick up a new value, restart the deployment.

Available at build and at runtime

Available at build and at runtime

A binding reaches the worker twice:

  • At runtime, as an environment variable, read when the worker boots.
  • At build time, as a Docker build secret with the same name. Mount it in a RUN step, for example to install packages from a private registry:
RUN --mount=type=secret,id=UV_INDEX_USERNAME,env=UV_INDEX_USERNAME \
    --mount=type=secret,id=UV_INDEX_PASSWORD,env=UV_INDEX_PASSWORD \
    uv sync --frozen --no-dev

The secret is available to that RUN step only and never persists in the image's history or layers. For compatibility, the value is also passed as a build arg with the same name, so existing Dockerfiles keep building. But a matching ARG writes the value into the image's history, and the build log warns about it: drop the ARG and read the secret with the mount instead.