Summary
On July 22, 2026, OpenAI launched Presence, an enterprise platform for building and running realtime voice and chat agents with policies, permission controls, escalation rules, pre-deployment simulations, and a Codex-powered loop that proposes improvements for human approval. It integrates GPT-Live for realtime voice and is delivered via OpenAI Forward Deployed Engineers rather than self-serve, with early users including BBVA Mexico and SoftBank.
What changed
OpenAI released Presence, a deployed enterprise product bundling company policies, access permissions, human-escalation rules, pre-deployment simulations, and Codex-driven post-launch improvement for realtime voice and chat agents.
Why it matters
Presence moves OpenAI from selling models and APIs to selling a governed agent deployment framework for regulated enterprise workflows like support, sales, procurement, and IT, competing directly with contact-center and agent-ops platforms and raising the bar on built-in guardrails, evaluations, and human-in-the-loop control.
Evidence excerpt
Presence bundles ... company policies and standard operating procedures, permission controls ... escalation rules ... pre-deployment simulations ... OpenAI's Codex agent reviews interactions ... but staff members must test and approve changes before they go live.