Stages

Learn about defining stages for runtime protection to be used during different stages in your application workflow

Stages allow you to map to and control runtime protection at different stages in your applications workflow. For example, you can have a stage to run rules against the input from a user, and a different stage for the output from an agentic workflow.

A stage contains a prioritized collection of rulesets. A stage is triggered if any of the rulesets are triggered, and the action returned is the action from the highest priority ruleset that is triggered. For example, if a stage has four rulesets in the order 1, 2, 3, and 4, and 2 and 4 are triggered, then the action from 2 would be returned.

Stage concepts

Stages are created in code against a project, then used in runtime protection by name or Id. Stages can have one of two types, central stages or local stages. You can provide the project name or project Id directly when creating a stage, or use the GALILEO_PROJECT environment variable.

When you create a central stage, you need to provide a prioritized list of rulesets. The order of these rulesets determines the action that is returned, if any rulesets are triggered, the action from the first triggered ruleset in the list is returned.

Stages can be paused and resumed when required. When paused, the stage will always return success (with a status of paused) with no rulesets triggered.

Stages can also be versioned, with different versions having different rulesets. When you use a stage, you can specify the version to use.

Central stages

Central stages are designed to be created and managed by central AI governance or IT teams, and can be used by any application. When you create a central stage, you supply the rulesets at creation time. When these stages are updated, for example adding new rulesets, the version is incremented. The applications using these stages can then either use a stage with a fixed version, or use the latest version.

Once a central stage has been created, it can be used in any application that uses the same Galileo project, using the stage Id or name.

Local stages

Local stages are designed to be managed by application teams. These stages are created without rules, then when your application uses the stage, it supplies the ruleset at runtime.

Local stages can be re-used between applications, but each application will need to provide the rulesets at runtime.

Create stages

Stages need to be created before being used, but can be created at any time. For central stages, these can be created using scripts managed by central AI governance teams. For local stages, you can create these in your application code (checking to see if they exist first), or create manually using scripts or part of your deployment pipelines.

See the create_protect_stage Python SDK docs for more details.

Create a central stage

When you create a central stage, you need to provide the prioritized list of rulesets.

Python

from galileo import GalileoMetrics
from galileo.stages import create_protect_stage

from galileo_core.schemas.protect.rule import Rule, RuleOperator
from galileo_core.schemas.protect.ruleset import Ruleset
from galileo_core.schemas.protect.stage import StageType

# Create a rule
rule = Rule(
    metric=GalileoMetrics.input_toxicity,
    operator=RuleOperator.gt,
    target_value=0.1
)

# Add this rule to a ruleset, using the default passthrough action
ruleset = Ruleset(rules=[rule])

# Create a stage
stage = create_protect_stage(
    name="My stage",
    stage_type=StageType.central,
    prioritized_rulesets=[ruleset],
    description="Test the input for toxicity."
)

Create a local stage

When you create a local stage, you don’t provide rulesets up front.

Python

from galileo.stages import create_protect_stage

from galileo_core.schemas.protect.stage import StageType

# Create a stage
stage = create_protect_stage(
    name="My stage",
    stage_type=StageType.local,
    description="A local stage with rules set when it is used"
)

Get stages

You can get a stage by name or Id.

Python

from galileo.stages import get_protect_stage

stage = get_protect_stage(stage_name="My stage")

If the stage doesn’t exist, this returns None.

See the get_protect_stage Python SDK docs for more details.

Update central stages

You can update the rulesets associated with a central stage. The stage can be updated to have a new ruleset, or you can clear the rules for a stage. When you update a stage, a new version is created.

Python

from galileo import GalileoMetrics
from galileo.stages import update_protect_stage

from galileo_core.schemas.protect.rule import Rule, RuleOperator
from galileo_core.schemas.protect.ruleset import Ruleset
from galileo_core.schemas.protect.stage import StageType

# Create a rule
rule = Rule(
    metric=GalileoMetrics.input_toxicity,
    operator=RuleOperator.gt,
    target_value=0.1
)

# Create a ruleset
ruleset = Ruleset(rules=[rule])

# update the stage
stage = update_protect_stage(
    stage_name="My stage",
    prioritized_rulesets=[ruleset],
)

See the update_protect_stage Python SDK docs for more details.

Pause and resume stages

Stages can be paused and resumed. This allows you to turn stages on and off without re-deploying your application.

pause a stage

Python

from galileo.stages import pause_protect_stage

# Pause a stage
pause_protect_stage(stage_name="My stage")

See the pause_protect_stage Python SDK docs for more details.

Resume a stage

Python

from galileo.stages import resume_protect_stage

# Resume the stage
resume_protect_stage(stage_name="My stage")

See the resume_protect_stage Python SDK docs for more details.

Related resources

Runtime protection basics

Learn the basics of running runtime protection.

Rules

Learn about defining rules for runtime protection.

Rulesets

Learn about defining rulesets for runtime protection.

Invoke runtime protection

Learn how to invoke runtime protection in code using the Galileo SDK.