Realtime voice over Twilio Streams, two-way SMS via Twilio Messaging, and inbound/outbound email through SendGrid. Every interaction is persisted with transcripts, metadata, and guardrail state, then forwarded to analytics and billing—no additional middleware required.

Virtual employees are fully configurable: set role, objectives, tone, language hints, onboarding/closing scripts, allowed and forbidden topics, regex blocks, toxicity thresholds, escalation rules, and maximum turns. Those guardrail settings are stored per assistant so voice, SMS, and email inherit them automatically.

We ship managed defaults, and you can bring your own vendor config for OpenAI, Azure, Google, or a custom endpoint. Vendor settings live per user in the User Management service and are applied at runtime, with guardrails wrapping each request before a response is delivered.

Integrations are API-first today. Communication webhooks, analytics endpoints, and billing usage feeds expose transcripts, guardrail events, and usage counters so teams can wire data into CRM or HR systems via REST or their integration platform. Tell us which packaged connectors matter most and we will prioritize them.

Guardrails run on every channel to enforce allowed topics, regex/PII rules, toxicity thresholds, override messaging, and escalation counts. Violations can mark a conversation for human review, are recorded in analytics, and every service enforces API keys, RS256 JWTs, and Twilio/SendGrid signature checks to keep traffic policy aligned.

Analytics covers conversation summaries, intent, sentiment by speaker, frustration trends, dominant emotions, engagement metrics, language quality (LanguageTool/Textstat), topics, actionable insights, and guardrail flags. Results live in MongoDB and are accessible via API for dashboards or downstream coaching.

Access tokens are RS256-signed JWTs with issuer/audience validation, backed by refresh-token revocation in Redis and route-level rate limiting. Internal APIs also require service keys, and channel webhooks verify Twilio and SendGrid signatures while IP and WebSocket throttles protect realtime traffic.

Plans define seat counts and included call, text, and email events for each vendor mode. The communication service emits a usage event per interaction (including vendor flag and quantity), the billing service aggregates counts, applies overage pricing, and issues invoices through Stripe.

Usage is tallied as discrete events: each completed voice call counts as one unit (or a custom quantity you supply), each SMS message increments text usage, and each email message increments email usage. Those counts feed dashboards and invoices alongside the included allowances in your plan.

Yes. We generate Stripe Billing Portal sessions so admins can upgrade, downgrade, or cancel; Stripe controls effective dates and we sync subscription state through webhooks so User Management and Billing stay aligned.

Conversation data—including transcripts, guardrail state, objectives, escalation flags, and vendor metadata—lives in MongoDB with shared IDs across services. Analytics enrichments and usage events reference the same IDs, access is gated by API keys and JWTs, and sensitive values are redacted in logs for operational visibility without leakage.

Most teams create an AI assistant, choose a model provider, and start testing calls or messages the same day. From there, add guardrails and analytics as you scale to production.