Documentation menu

Get Started

CLI Usage

Every adapter has a CLI entry point. Pass a scenario and ARIA runs the conversation, scores it, and writes a report — no dashboard required.

Commands

Each command targets a different agent platform:

CommandTargets
npm run cli:connectAmazon Connect (chat / voice)
npm run cli:lexAWS Lex V2
npm run cli:azureAzure Direct Line
npm run cli:azure-openaiAzure OpenAI Agents Runtime
npm run cli:strandsStrands / AgentCore HTTP
npm run cli:copilotGitHub Copilot Chat
npm run cli:openapiAny OpenAPI 3.x HTTP endpoint
npm run cli:customGeneric HTTP / WebSocket endpoint

Running a scenario

Pass arguments after -- so npm forwards them to the CLI:

npm run cli:openapi -- --scenarios-dir=./scenarios --scenario=banking/functional/account_query

Each judge call prints its token usage as [Xin/Yout] in the terminal, and the run writes a paired HTML + JSON report when it finishes.

Common flags

FlagDescription
--scenario, -sPath-prefix filter for scenarios to run (relative to the scenarios dir), e.g. banking/functional/account_query.
--scenarios-dirDirectory to load scenarios from. Use ./scenarios for the repo’s bundled packs.
--channel, -cchat (default) or voice.
--transcript, -tRe-evaluate a saved transcript JSON without re-running the conversation.
--conversation-only / --no-evalRun the conversation without invoking the judge.

New to scenarios? See Scenarios for the YAML format.