Brilo AI Sandbox Mode lets you test knowledge changes, training data, and call flows in an isolated test environment before those changes go live. A separate knowledge set and test routing are used so the Brilo AI voice agent answers test calls, runs validation checks, and logs results without affecting production calls or live customers. Use Sandbox Mode to run dry runs, confirm intent mapping, and validate webhooks or integrations before promoting changes to production. Sandbox Mode supports versioning and rollback workflows so you can compare outcomes between the sandbox and production voice agents.

Can I test knowledge updates in a sandbox?

Yes — Brilo AI Sandbox Mode runs your knowledge updates against a non-production voice agent so you can validate behavior before deployment.

How do I run test calls for knowledge changes?

Run test calls from the Brilo AI dashboard or use a test webhook to exercise the sandboxed knowledge set; review transcripts and intent confidence scores to validate results.

Can I validate integrations and webhooks in sandbox?

Yes — when configured, Sandbox Mode will call your webhook endpoint and simulate integration responses without touching production routing.

Can I revert a knowledge change after testing?

Yes — Sandbox Mode includes versioning so you can keep the previous production knowledge set intact and roll back if testing surfaces issues.

Buyers ask about Sandbox Mode because changing knowledge or retraining can unintentionally degrade live call quality. Enterprises in healthcare, banking, and insurance need a safe way to validate updates without risking customer experience or regulatory exposure. Testing knowledge changes, intent updates, and new dialogue paths in an isolated environment reduces change-related incidents and supports controlled releases and compliance reviews.

Brilo AI Sandbox Mode creates an isolated copy of your voice agent’s knowledge base, dialog flows, and routing rules for testing. When enabled, the Brilo AI voice agent will route calls or simulated sessions to the sandboxed environment instead of production, letting you confirm responses, intent classification, and webhook behavior. Typical sandbox activities include running representative test calls, validating answer quality, and checking fallback and escalation conditions.

A sandbox knowledge set is a copy of your live knowledge base used only for testing and not served to real customers. Consider running sample test calls and accent/locale variations to ensure voice and ASR performance before promoting changes.

Brilo AI Sandbox Mode is intentionally limited to testing. It should not be used for production traffic, customer-facing updates, or any activity that requires audit-level production logging unless explicitly configured. Sandbox sessions may not include the same data retention, monitoring integrations, or third-party connectors as production by default. Before using sandbox test data for regulatory purposes, confirm your organization’s compliance requirements and log policies.

In Brilo AI, a training run (dry run) is a non-production execution of updated training data that collects metrics and example transcripts without modifying the production model. Do not assume sandbox results will exactly match production performance; differences can occur because production monitoring, scaling, or certain integrations may behave differently.

For fallback and uncertain answers, see Brilo AI’s guidance on how the system behaves when unsure: Brilo AI — What happens when the AI is unsure?

When a sandbox test reveals an escalation condition, Brilo AI can simulate the human handoff path without connecting to live agents. Configure sandbox routing so that escalations either record a simulated transfer or send a non-production webhook to your test endpoint. In production, the Brilo AI voice agent uses the same escalation rules but routes to live queues or contact centers; Sandbox Mode keeps those handoffs contained for validation and training. Use the test logs to confirm that handoff prompts, context passed to agents, and metadata (for example, intent and confidence) are correct before enabling them in production.

For more on running representative test calls and adjusting language/locale, see: Brilo AI — How does the AI handle accents and speech variations?

Using Brilo AI Sandbox Mode reduces deployment risk by catching regressions before they reach customers. Buyers can expect fewer live-call incidents, clearer validation workflows for compliance reviews, and faster, safer iteration on knowledge content. Sandbox testing supports staged rollouts and improves stakeholder confidence in production changes without claiming specific uptime or compliance guarantees.

Can Sandbox Mode use real customer data?

No. Sandbox Mode should not be used with live customer data unless you have explicit legal and compliance approval. Use synthetic or anonymized test data for safe validation.

How long does a sandbox copy persist?

Sandbox retention varies by your Brilo AI plan and account settings. Consult your Brilo AI admin settings to confirm sandbox lifecycle policies before creating long-running test environments.

Will sandbox tests affect production analytics?

Sandbox tests are isolated and do not feed into production analytics unless you explicitly configure a shared reporting endpoint. Keep reporting separated to avoid polluting production metrics.

Can I promote a sandbox knowledge set to production?

Yes — after validation, you can promote tested knowledge changes to production using the Brilo AI promotion workflow or by exporting and applying the validated knowledge set to the production voice agent.

Do sandbox tests include webhook and CRM integrations?

Sandbox Mode can call your test webhook endpoint and simulate CRM updates when configured, but it will not modify your production CRM unless you point it to production integration endpoints.