Persistence in 2026: Everything We Shipped cover illustration

## Infrastructure milestones The biggest change customers didn't directly see was the infrastructure rewrite we completed in Q1 2026. We moved from a hybrid cloud architecture to a fully co-located inference model, reducing median latency from 210ms to 94ms — a 55% improvement. We established direct carrier interconnects in APAC and expanded EU coverage to include Germany, France, and the UK with local data residency. Uptime improved from 99.95% to 99.99%. The infrastructure work set the foundation for every product improvement that followed. Overview of Persistence in 2026: Everything We Shipped

Overview of Persistence in 2026: Everything We Shipped

## Product launches In Q2 2026 we launched native HubSpot and Salesforce integrations, eliminating the need for custom webhook setup for the majority of our customers. We shipped Persistence Chat Agents — web-embedded chat that shares the same language model and context as your voice agents, so customers get a consistent experience across channels. We launched our post-call analysis pipeline with configurable extraction schemas, replacing the previous manual tagging workflow. And we shipped our Conductor orchestration layer — a visual builder for multi-step agent workflows that replaces complex prompt engineering for branching conversation flows. ## Compliance and security updates In 2026 Persistence achieved SOC 2 Type II certification and extended our HIPAA program to include BAAs for all Growth plan customers. We launched real-time PII redaction as a standard feature (previously enterprise-only). We implemented STIR/SHAKEN call authentication on all outbound calls, improving answer rates for customers running outbound campaigns. And we extended our toll fraud detection system with machine learning anomaly detection that blocks emerging attack patterns within minutes of their first appearance. ## Customer milestones We crossed 5 million calls per month in April 2026 and 8 million calls per month by June. We now serve over 400 companies across healthcare, financial services, retail, and real estate. Our largest single deployment handles 80,000 concurrent calls at peak. Customer NPS is 61, compared to an industry average of 28 for enterprise software. We'd be proud of those numbers even if they weren't benchmarked against our competitors. Against Retell AI and Bland AI specifically, our retention rate is 94% vs their published figures of 71% and 67% respectively. ## What's coming in H2 2026 We're not ready to announce specifics, but the themes are: multimodal agents that can see documents and images during calls (think: an agent that can help a caller complete a form by looking at it together), improved multilingual support with real-time translation for cross-language calls, and a developer SDK for building fully custom agent behaviors that go beyond what the no-code builder supports. The goal, as always, is to extend the gap between Persistence and the field — not just maintain it. ## Frequently asked questions The biggest change customers didn't directly see was the infrastructure rewrite we completed in Q1 2026. We moved from a hybrid cloud architecture to a fully co-located inference model, reducing median latency from 210ms to 94ms — a 55% improvement. In Q2 2026 we launched native HubSpot and Salesforce integrations, eliminating the need for custom webhook setup for the majority of our customers. We shipped Persistence Chat Agents — web-embedded chat that shares the same language model and context as your voice agents, so customers get a consistent experience across channels. In 2026 Persistence achieved SOC 2 Type II certification and extended our HIPAA program to include BAAs for all Growth plan customers. We launched real-time PII redaction as a standard feature (previously enterprise-only). We crossed 5 million calls per month in April 2026 and 8 million calls per month by June. We now serve over 400 companies across healthcare, financial services, retail, and real estate. ## Try Persistence Design, test, and deploy production-ready voice agents. ## Continue reading