Overview
An AI app is the workspace that groups all evaluation runs for a given AI use case, their results and artefacts and the tooling to govern the AI app, such as policies and risk scorers.
In the CLI, AI apps behave similarly to Git branches - an AI app needs to be created and switched to before running new evaluation runs. An AI app is identified by its key (cf. branch name).
Create and switch
An app is declared like any other entity. Beyond identity, risk_profile records the deployment context of the use case - it documents the app rather than changing how it is evaluated, and it is what governance reporting draws on:
app.yaml
key: "airline-assistant"
display_name: "Airline Virtual Assistant"
description: "Customer-facing assistant for flight bookings and changes."
long_description: !include "./README.md"
tags: ["customer-facing", "chat"]
risk_profile:
deployment_reach: "external_authenticated" # internal_restricted, internal_broad, external_authenticated, external_public
autonomy: "human_on_the_loop" # human_in_the_loop, human_on_the_loop, fully_autonomous
data_sensitivity: "personal_data_pii" # public_data_only, internal_business_data, personal_data_pii, special_category_data
action_capability: "reversible_write_actions" # generates_text_advice_only, read_only_tool_access, reversible_write_actions, irreversible_or_high_impact_write_actions
untrusted_content_exposure: "user_uploaded_or_third_party_documents" # direct_user_input_only, curated_internal_knowledge_base, user_uploaded_or_third_party_documents, live_web_email_or_peer_agent_content
domain_criticality: "business_operations" # general_information, business_operations, consequential_decisions, safety_critical
audience_vulnerability: "general_public" # internal_experts_trained_operators, professional_users, general_public, vulnerable_groupsCreate it, then switch to it. All evaluation runs will be executed inside that app, until you switch again:
lf add app -f app.yaml
lf switch airline-assistant
lf list applf init scaffolds this file for you, and lf status shows which app is currently active. Every property is documented in the AI apps reference.