Vijil client. While lifecycle methods coordinate complete workflows, resources let you start an operation without waiting, retrieve individual records, or use operations that do not have a lifecycle shortcut.
Standard list() methods return a Page[T]; iterate page.items for the results. Red Team artifact methods return their own typed responses.
client.agents
Manage Agent configurations.
archive() performs a soft deletion and returns the archived Agent. The older delete() method calls archive() for backward compatibility but returns None; prefer archive() in new code.
client.evaluations
Manage both Evaluation types. Use client.evaluate() for the polling workflow; create() starts an evaluation without waiting.
Evaluation with an ID. Use that evaluation_id for all Red Team status, artifact, and report reads. poll_status() is for behavioral evaluations; use redteam_status() for Red Team. Red Team does not accept baseline, Harness, or sample_size settings.
Prompt-Level Behavioral Responses
Useresponses() to retrieve flattened prompt and response rows from a completed behavioral evaluation, including an evaluation that uses a custom Harness. Each item identifies its parent Harness, Scenario, and Probe and includes the Agent response and raw Detector results.
Page[EvaluationResponse]. page.total is the number of rows that match the filters before pagination.
failed means that the parent Probe scored below the evaluation’s pass threshold. It does not describe an individual Detector result. Call responses() after the Console saves the evaluation results. The method is not available for Red Team evaluations; use the Red Team artifact methods instead.
Red Team Status, Artifacts, and Reports
All methods below take anevaluation_id and an optional team_id.
For example, inspect an attack and its judgment after the run completes:
RedteamStatusResponse for progress fields and terminal statuses.
client.harnesses
List and manage Harnesses. list() combines standard and custom Harnesses.
create() generates a custom Harness for an Agent and returns immediately by default. Pass wait=True to wait until generation reaches succeeded, failed, or cancelled. show() exposes the current generation status through harness.workflow.
Custom Harnesses are immutable. Calling update() raises NotImplementedError; create a replacement and delete the old Harness instead.
client.scores
Retrieve the latest Trust Score for an Agent, the score for a specific Evaluation, score history, or the full nested Evaluation results.
show() raises VijilNotFoundError when the Agent has no scores. See Understand Evaluation Results for guidance on interpreting scores.
client.reports
List, fetch, and download Trust Reports. download() returns PDF bytes. For a Red Team report, use client.evaluations.redteam_report(), redteam_report_markdown(), or redteam_report_pdf().
team_id argument, client configuration, or authenticated token. download() requests generation once if the PDF does not exist, then retries the download.
client.dome
Read and update Dome Guardrail configurations. Use client.protect() for the high-level configuration workflow.
client.jobs
Track and control asynchronous evolution jobs.
status() is an alias for show().
client.genomes
Version and compare Agent source across adaptations.
client.proposals
Review and apply the adaptation proposals produced by client.adapt().
client.policies
Manage the Policies that constrain Agent behavior. category must be privacy, ethics, security, compliance, operational, brand, or custom.
source_text or content. You can also pass file_path to read content from a UTF-8 text file.
extract_rules() creates draft rules from the policy source text and waits for extraction by default. Pass wait=False to return after starting extraction. rules() returns a paginated list and accepts an optional status filter.
client.personas
Manage the Personas used to build custom Harnesses. Pass role when creating a Persona. When provided, intent must be benign, curious, adversarial, or malicious.
client.monitor
Read available Dome runtime telemetry. The since argument accepts durations such as "1h", "24h", or "7d".
client.agent_tunnel
Provision or revoke the credential used to connect Vijil to a local Agent. get() creates the credential on its first call and returns the existing credential on later calls.
credential.token as a secret. After revoke(), the next get() call provisions a new credential.
client.rca_runs
Run root-cause analysis over a behavioral Evaluation’s failures and retrieve the resulting engineering tickets.
create() starts an asynchronous run and returns its ID and initial status. Poll show() until the status is completed or failed before requesting tickets. source_ref accepts a GitHub or S3 source reference.