If the phone rings, we answer. Even when something breaks.
An agent that is down is worse than no agent at all — it silently swallows the calls you were already paying for. Redundancy is not an upsell here; it is the architecture.
Four nodes, one private network
Redundant application tier, isolated data tier, and a load balancer that knows how to drain a node without hanging up on your customer.
Load balancer
Nginx terminates TLS, health-checks both app nodes, and drains traffic from an unhealthy node without dropping a live call.
- TLS 1.3
- Health checks
- Rate limiting
Application node A
Runs the realtime media pipeline, agent orchestration, and integration workers. Sized to carry full production load alone.
- Realtime media
- Agent runtime
- Webhooks
Application node B
An identical clone of the primary, serving live traffic in active-active and ready to absorb 100% of it during deploys or failure.
- Active-active
- Zero-downtime deploys
- Instant failover
Database & observability
PostgreSQL with point-in-time recovery, plus the metrics, log aggregation, and intrusion-detection agent for the whole fleet.
- PostgreSQL
- PITR backups
- IDS + metrics
Every hop between nodes travels over a private WireGuard mesh. No service port is exposed to the public internet except the load balancer, and database traffic never leaves the encrypted tunnel.
Built for the 2am no-heat call
The value of an always-on agent collapses the moment it is not on. These are the properties that keep it up.
No single point of failure
Two identical application nodes serve traffic behind a health-checking load balancer. Lose one mid-call and the other carries the load without a dropped session.
Private encrypted mesh
Application, database, and monitoring nodes talk only over WireGuard. The database has no public interface and no inbound rule that permits one.
Measured, not assumed
Answer latency, turn latency, transfer success and booking rate are tracked per call. Regressions page us before a customer notices.
Hardened by default
Key-only SSH, host firewalls that default-deny, automatic security patching, and a monitoring agent watching for intrusion signatures across the fleet.
- Target uptime
- 99.95%
- Failover time
- < 10s
- Backup cadence
- Continuous PITR
- Restore drill
- Quarterly
Your customers' conversations, handled like they matter
Recorded calls are some of the most sensitive data a business holds. We treat them that way, and we will walk your security reviewer through every control.
- Encryption in transit (TLS 1.3) and at rest (AES-256)
- Role-based access control with least-privilege service accounts
- Immutable audit log of every call, message, and configuration change
- Configurable retention — from indefinite down to 24-hour purge
- PII and PHI redaction available on transcripts and recordings
- BAA available on Enterprise for covered entities
- Signed webhooks with replay protection
- Annual third-party penetration test
Hear your own agent answer a live call in 20 minutes.
Bring a real scenario from your business. We will configure an agent on the call, dial it together, and you decide whether it beats what happens today.
No slide deck. No procurement gauntlet. Live in 7 days.