Lifecycle Methods

Lifecycle methods are high-level workflows on the Vijil client. Each method maps to a stage of the Agent lifecycle. Long-running operations poll until they reach a terminal state unless the method provides a no_wait option.

Method	What It Does	Returns
`client.evaluate()`	Run a behavioral or red-team Evaluation	`Evaluation \| RedteamStatusResponse`
`client.adapt()`	Improve an Agent through corrective evolution	`Job`
`client.protect()`	Configure Dome runtime Guardrails	`DomeConfig`
`client.discover()`	Find Agents in GitHub repositories or cloud infrastructure	`Page[DiscoveredAgent]`
`client.register()`	Register Agent source from a local directory	`dict[str, Any]`
`client.evolve()`	Create an Agent through generative evolution	`Job`
`client.deploy()`	Deploy an Agent to a production runtime	`Deployment`

`client.evaluate()`

Run an Evaluation and poll until it reaches a terminal state. The default type="behavioral" path uses Harnesses and returns an Evaluation with the finalized scores.

evaluation = client.evaluate(
    "agent-id",
    baseline=True,
    sample_size=50,
)

print(evaluation.status)
print(evaluation.trust_score)
print(evaluation.dimensions)

Parameter	Type	Default	Description
`agent_id`	`str`	—	Agent ID or alias
`baseline`	`bool`	`False`	Run the standard reliability, security, and safety Harnesses
`harness_id`	`str \| None`	`None`	Run a specific Harness instead of the baseline
`sample_size`	`int \| None`	`None`	Number of Probes to run per Harness; must be at least `1`
`type`	`str`	`"behavioral"`	Evaluation type: `"behavioral"` or `"redteam"`
`redteam`	`Mapping[str, Any] \| None`	`None`	Red-team configuration overrides, such as `max_waves` or `advanced`
`_poll_interval`	`float`	`5.0`	Seconds between status checks
`on_poll`	`Callable \| None`	`None`	Callback invoked with the current state after each status check

For a behavioral Evaluation, pass either baseline=True or harness_id, not both. To start a behavioral Evaluation without waiting, use client.evaluations.create() and poll the Evaluation resource yourself. See Run Evaluations for the lower-level workflow.

Run a Red-Team Evaluation

Set type="redteam" to challenge an Agent with adaptive adversarial waves. This replaces the removed client.test() workflow.

status = client.evaluate(
    "agent-id",
    type="redteam",
    redteam={"max_waves": 3},
)

print(status.evaluation_id)
print(status.status)
print(status.current_wave)

This path returns a RedteamStatusResponse and polls until the status is succeeded, failed, cancelled, or degraded. Do not combine type="redteam" with baseline, harness_id, or sample_size; the SDK rejects those behavioral-only arguments.

`client.adapt()`

Create a corrective evolution job for an existing Agent. By default, the method polls until the job reaches a terminal status: completed, failed, or cancelled.

job = client.adapt("agent-id", mode="config")
print(job.status)

Parameter	Type	Default	Description
`agent_id`	`str`	—	Agent ID or alias
`mode`	`str`	`"config"`	`"config"` for GEPA or `"code"` for ADAS/SourceEngine
`no_wait`	`bool`	`False`	Return the initial `Job` without polling

Returns a Job. After adaptation completes, review and apply the resulting proposals.

`client.protect()`

Configure Dome runtime Guardrails for an Agent.

dome = client.protect(
    "agent-id",
    guards=["prompt_injection", "pii"],
    mode="enforce",
)

print(dome.mode)
print(dome.guards)

Parameter	Type	Default	Description
`agent_id`	`str`	—	Agent ID or alias
`policy_id`	`str \| None`	`None`	Policy to apply
`guards`	`list[str] \| None`	`None`	Guard names
`mode`	`str`	`"enforce"`	`"enforce"` blocks threats, `"monitor"` logs them, and `"disabled"` turns Guardrails off

Returns a DomeConfig. For local Dome configuration, see Configure Guardrails.

`client.discover()`

Find Agents in GitHub repositories or supported cloud infrastructure. Provide at least one GitHub selector or a provider.

# Discover repositories in a GitHub organization.
github_agents = client.discover(github_org="acme-corp")

# Discover Agents in a Kubernetes namespace.
cloud_agents = client.discover(
    provider="k8s",
    namespace="production",
)

Parameter	Type	Default	Description
`github_org`	`str \| None`	`None`	GitHub organization to scan
`github_repo`	`str \| None`	`None`	GitHub repository to scan
`provider`	`str \| None`	`None`	Infrastructure provider, such as `"aws"` or `"k8s"`
`region`	`str \| None`	`None`	Cloud region filter used with `provider`
`namespace`	`str \| None`	`None`	Kubernetes namespace filter used with `provider`

Returns a Page[DiscoveredAgent].

`client.register()`

Register Agent source from a local directory. The SDK uploads text source files to the Console import route and returns the response as a dictionary.

result = client.register("./my-agent")
print(result["id"])

Parameter	Type	Default	Description
`path_or_url`	`str`	—	Path to a local directory containing the Agent source
`discover_id`	`str \| None`	`None`	Reserved parameter; discovery ID registration is not supported

Despite the parameter name, path_or_url does not accept a GitHub URL. Clone the repository first, then pass its local directory. The SDK excludes common dependency, cache, version-control, binary, and secret files. It rejects source trees larger than 5 MiB or 500 files instead of uploading a partial tree.

`client.evolve()`

Create an Agent from a specification, a local YAML or JSON specification file, or a natural-language description. The method polls the evolution job unless you set no_wait=True.

job = client.evolve(
    description="A travel booking Agent for flights and hotels",
)
print(job.status)

Parameter	Type	Default	Description
`spec`	`dict[str, Any] \| None`	`None`	Evolution specification
`spec_path`	`str \| None`	`None`	Path to a YAML or JSON specification file
`description`	`str \| None`	`None`	Natural-language description lowered to a specification
`no_wait`	`bool`	`False`	Return the initial `Job` without polling

Provide one of spec, spec_path, or description. Returns a Job.

`client.deploy()`

Deploy an Agent to a production runtime. The method polls while the deployment status is deploying unless you set no_wait=True.

deployment = client.deploy(
    "agent-id",
    runtime="agentcore",
)

print(deployment.status)
print(deployment.url)

Parameter	Type	Default	Description
`agent_id`	`str`	—	Agent ID or alias
`runtime`	`str`	—	Target runtime, such as `"agentcore"`, `"do-gradient"`, or `"k8s"`
`namespace`	`str \| None`	`None`	Kubernetes namespace used with the `"k8s"` runtime
`no_wait`	`bool`	`False`	Return the initial `Deployment` without polling

Returns a Deployment. Discovery, evolution, and deployment availability depends on the capabilities enabled in your Console deployment.

Next Steps

SDK Resources

Use lower-level resource operations and pagination.

Models and Errors

Inspect return types and handle SDK exceptions.

Run Evaluations

Start Evaluations and retrieve detailed results.

Protect Agents

Add runtime Guardrails with Dome.

Get Started

Evaluate Agents

Red Team

Protect Agents

Reference

Lifecycle Methods

`client.evaluate()`

Run a Red-Team Evaluation

`client.adapt()`

`client.protect()`

`client.discover()`

`client.register()`

`client.evolve()`

`client.deploy()`

Next Steps

SDK Resources

Models and Errors

Run Evaluations

Protect Agents

​client.evaluate()

​Run a Red-Team Evaluation

​client.adapt()

​client.protect()

​client.discover()

​client.register()

​client.evolve()

​client.deploy()

​Next Steps

SDK Resources

Models and Errors

Run Evaluations

Protect Agents

`client.evaluate()`

Run a Red-Team Evaluation

`client.adapt()`

`client.protect()`

`client.discover()`

`client.register()`

`client.evolve()`

`client.deploy()`

Next Steps