Skip to main content
Blog

The Orbit blog

Field notes on AI-era growth, deliverability, voice, and messaging for developers and teams building on Orbit by Devotel.

  • Orbit Editorial Team

    The PII egress guard for AI agents, explained

    The failure mode that worries security reviewers is not a hallucinated answer. It is a tool call that succeeds with a real phone number, email, or identifier its operator never intended to send. This explainer covers Devotel Orbit's opt-in safety_config.pii_egress flag, the detector floor it scrubs before dispatch, the honest limit for tools that parse E.164, the 400/422 write contract, and how the guard composes with human-in-the-loop approvals.

    • AI agents
    • PII
    • data minimization
    • guardrails
    • tool calls
    • agent safety
  • Orbit Editorial Team

    Measuring guardrail effectiveness — the operator's guide to the Guardrail Effectiveness surface

    Configuring a guardrail and knowing it fires on live traffic are different claims. The Guardrail Effectiveness dashboard in Devotel Orbit closes the gap — refusal and block rates, violation trends over a rolling window, and per-policy firing counts that name the rule. This guide walks the metrics panel, the diagnostic patterns (silent policies, over-firing policies), and the full configure → observe → adjust loop.

    • AI agents
    • guardrails
    • observability
    • monitoring
    • operator guide
  • Orbit Editorial Team

    Eval guardrails vs generic AI-voice quality checks — why a rubric alone cannot stop a bad call

    Generic AI-voice quality checks score conversations against a rubric and stop there. Eval guardrails close the loop with tenant-authored rule dictionaries, per-agent policy versions, deterministic production sampling, pre-promotion replay, and analytics that show which rule fired. This comparison shows what each ships and what a rubric-only program cannot catch.

    • AI voice agents
    • guardrails
    • evaluation
    • quality assurance
    • comparison
  • Orbit Editorial Team

    Orbit Eval Guardrails vs Generic AI-Voice Quality Checks: The Sad-Path Playbook Buyers Should Grade Vendors On

    Generic AI-voice quality checks grade the average call and stop. Eval guardrails grade the sad paths, the failure modes production traffic finds first. This comparison defines the six checks a quality program needs, maps each to Devotel Orbit's shipped surface, and holds the vendor field to the same rubric.

    • comparison
    • AI voice agents
    • evaluation
    • guardrails
    • quality assurance
    • voice of customer
  • Orbit Editorial Team

    Human-in-the-loop tool approvals for AI agents

    Every tool an AI agent calls is either unguarded or gated, and the gate either pauses the call before it runs or merely reviews it after. This guide walks the human-in-the-loop approval queue on Devotel Orbit — synchronous pre-execution gating, the supervisor decision loop, cost previews on the pause card, and the audit trail every decision writes.

    • human-in-the-loop
    • tool approvals
    • AI agents
    • agent safety
    • supervisor review
    • guardrails
  • Orbit Editorial Team

    Custom Guardrail DSL, Announced: Tenant-Authored Rules Beyond the Five Built-In Scanners

    The Devotel Orbit safety stack now closes the tenant-specific gap the fixed scanners cannot express. Author literal `contains` or regex rules on `safety_config.custom_guardrail_rules` with block, redact, or warn actions, evaluated deterministically on every agent turn and shipped today.

    • AI agents
    • guardrails
    • safety
    • announcement
    • 2026

Ready to build on Orbit?

Start free and ship your first voice, messaging, or email flow today, or talk to our team about your rollout. No annual contract, one pay-as-you-go bill across every channel.

Start free Talk to sales

See transparent pay-as-you-go pricing

Blog — Orbit by Devotel