vibium ready
Check the selected local browser executable files, then test AI when configured. No browser or driver is launched.
Synopsis
Section titled “Synopsis”Nightly only. This command is in the nightly builds but not yet in the latest npm release. See What’s in Nightly.
vibium ready [flags]vibium ready [command]AI Flags
Section titled “AI Flags”--ai-base-url string AI provider API base URL, not the site under test (env: VIBIUM_AI_BASE_URL); empty resets the provider default--model string Model ID for the provider; see Model IDs below (env: VIBIUM_AI_MODEL)--provider string AI provider: openai, xai, anthropic, google, openai-compatible, or local (env: VIBIUM_AI_PROVIDER); changing it requires --model and resets endpoint/effort defaults--reasoning-effort string none, minimal, low, medium, high, xhigh, or max; not for anthropic or google (env: VIBIUM_AI_REASONING_EFFORT); empty uses the model defaultModel IDs
Section titled “Model IDs”openai https://developers.openai.com/api/docs/modelsxai https://docs.x.ai/developers/modelsanthropic https://platform.claude.com/docs/en/models/overviewgoogle https://ai.google.dev/gemini-api/docs/modelsopenai-compatible the model name your server serveslocal the model name your server servesSubcommands
Section titled “Subcommands”vibium ready ai
Section titled “vibium ready ai”Require valid AI configuration and test authentication, model access, tool calling, and a structured response.
Nightly only. This command is in the nightly builds but not yet in the latest npm release. See What’s in Nightly.
vibium ready ai [provider] [flags]AI Flags
Section titled “AI Flags”--ai-base-url string AI provider API base URL, not the site under test (env: VIBIUM_AI_BASE_URL); empty resets the provider default--model string Model ID for the provider; see Model IDs below (env: VIBIUM_AI_MODEL)--provider string AI provider: openai, xai, anthropic, google, openai-compatible, or local (env: VIBIUM_AI_PROVIDER); changing it requires --model and resets endpoint/effort defaults--reasoning-effort string none, minimal, low, medium, high, xhigh, or max; not for anthropic or google (env: VIBIUM_AI_REASONING_EFFORT); empty uses the model defaultModel IDs
Section titled “Model IDs”openai https://developers.openai.com/api/docs/modelsxai https://docs.x.ai/developers/modelsanthropic https://platform.claude.com/docs/en/models/overviewgoogle https://ai.google.dev/gemini-api/docs/modelsopenai-compatible the model name your server serveslocal the model name your server servesvibium ready ai# Tests the configured provider and model.vibium ready ai anthropic --model your-model# Tests Anthropic with the supplied model and ANTHROPIC_API_KEY.vibium ready ai xai --model grok-4# Tests xAI with the supplied model and XAI_API_KEY.vibium ready ai --json# Prints the provider checks as JSON.vibium ready ai --provider local --model my-model --ai-base-url http://127.0.0.1:8080/v1 --reasoning-effort ""# Uses these settings for one invocation; credentials still come from the environment.vibium ready browser
Section titled “vibium ready browser”List discovered Vibium browser installations and check the selected Chrome or Firefox executable files.
Nightly only. This command is in the nightly builds but not yet in the latest npm release. See What’s in Nightly.
vibium ready browser [engine] [flags]vibium ready browser# Checks the default/selected installation; other discovered installations are informational.vibium ready browser firefox --channel beta# Checks Firefox beta installation without launching it or changing defaults.vibium ready browser chrome --json# Structured installation results; browser connection is reported as skipped.Examples
Section titled “Examples”vibium ready# Checks browser installation and configured AI; reports fixes or READY.vibium ready --json# Structured readiness results; exit 0 when requested checks pass, otherwise 1.vibium ready --provider local --model my-model --ai-base-url http://127.0.0.1:8080/v1 --reasoning-effort ""# Uses these settings for one invocation; credentials still come from the environment.