Operating ProcedureExecution layer

Failproof AI Client Agent Monitoring Setup (Onboarding)

A sequence with 7 steps: Create a Failproof AI account and select the Free Forever plan.

By InnovaAI ResearchPublished

sequence

Failproof AI Client Agent Monitoring Setup (Onboarding)

  1. 01

    Create a Failproof AI account and select the Free Forever plan

    Start at the signup page and choose the Free Forever plan ($0/month) to access the workspace. This plan includes 5,000 runs per month, 100 evals, and 3 failure audits, which is sufficient for initial testing.

  2. 02

    Install the Failproof CLI and connect your first agent harness

    In the workspace, navigate to the Integrations section. Install the Failproof CLI and connect your agent harness (e.g., Claude Code, Cursor, Codex, or Gemini CLI) by following the setup wizard. This captures runtime logs for tracing.

  3. 03

    Enable deep tracing for the connected agent

    In the agent settings, toggle on Deep Agent Tracing. This ensures every run is logged with detailed execution paths, which is essential for debugging client-facing agents.

  4. 04

    Configure custom failure audit policies

    Go to the Policies tab and start with the 39 built-in policies. Create custom policies that match your client's top failure scenarios, such as unsafe code generation or data leakage. Set the audit frequency to daily to stay within the free tier's 3 audits per month.

  5. 05

    Set up alerting thresholds and notifications

    In the Alerting suite, define thresholds for critical failures (e.g., silent failures, policy violations). Configure email or webhook notifications so your agency team is alerted immediately when a client agent misbehaves.

  6. 06

    Build a branded observability dashboard for the client

    Use the built-in dashboards to create a view showing agent health, run history, and failure trends. Customize the dashboard with your agency's branding to present a professional monitoring report to the client.

  7. 07

    Document the runbook and hand off to the client

    Write a runbook that explains how to read the dashboard, what alerts mean, and how to respond to common failures. Share this with the client and schedule a review meeting to walk them through the monitoring setup.