OpenAI Presence
Last edited
Fact-checked
In review queue
Sources
8 citations
Revision
v1 · 1,361 words
Fact-checks are independent of edits: a reviewer re-verifies the article against its sources and stamps the date. How we verify
OpenAI Presence is a managed enterprise platform from OpenAI for deploying governed AI agents in customer-facing and internal workflows. OpenAI announced the product on July 22, 2026, with support for real-time voice and chat experiences. It entered limited general availability for eligible enterprise customers as a deployed service rather than a self-serve software product.[1][2]
Presence combines OpenAI models with workflow integration, policies, permissions, testing, monitoring, and human escalation. The platform is intended for high-volume or high-stakes work in which an agent may retrieve information, update business systems, or take an approved action. OpenAI identifies customer support, outbound sales, insurance claims, billing, and employee IT service as early examples.[1][2]
Product scope and capabilities
A Presence deployment begins with a defined job. The agent receives approved knowledge and only the system access needed for that job. The customer specifies the instructions, standard operating procedures, permission boundaries, approval points, and conditions that require a handoff to a person or another support channel.[1][2]
| Area | Documented capability | Deployment boundary |
|---|---|---|
| Knowledge | Use approved company information for a defined workflow | Sources and access are selected during technical scoping |
| Business systems | Connect through APIs and tools to retrieve data, update records, or complete approved actions | Permissions are scoped to the job |
| Channels | Communicate through supported voice or chat experiences | Channel, routing, authentication, and contact-center integration vary by deployment |
| Escalation | Transfer work to a person or another support path when policy or risk requires judgment | Handoff rules and receiving systems are customer-specific |
| Operations | Record sessions and actions, monitor quality signals, and test revised versions | Logging, retention, access, and rollout procedures are defined for each deployment |
The platform is not an agent that becomes production-ready after documents are uploaded. OpenAI's deployment guidance calls for systems integration, policy encoding, simulations, evaluations, acceptance testing, and formal approval before a controlled launch.[2]
Managed deployment model
Presence deployments are led by OpenAI Forward Deployed Engineers, selected deployment partners, or both. The process starts by defining business outcomes and success criteria. The team then connects the required systems, encodes permissions and escalation paths, completes security and legal review, tests the agent, and stages a monitored production rollout. Later changes go through further review and testing.[2]
This delivery model fits OpenAI's wider investment in embedded enterprise engineering. In May 2026, the company launched the OpenAI Deployment Company to place Forward Deployed Engineers inside organizations working on production systems. Independent reporting described this approach as a move into implementation work that technology vendors have often left to consultants.[7][8]
Presence differs from ChatGPT Enterprise Workspace Agents. Workspace Agents are created and managed by users inside supported ChatGPT and Slack workspaces, where they can be shared, scheduled, or triggered through an API. Presence is scoped and deployed with OpenAI or a selected partner for production workflows that require deeper integration and operational support. It is not created through the Workspace Agent interface.[2][5]
Governance and improvement process
OpenAI describes governance as part of the product's operating system rather than a final review step. Before release, simulations and graders can test ordinary requests, edge cases, policy compliance, tool use, and escalation behavior. Guardrails, scoped permissions, and approval steps restrict the actions available to the agent. Teams can also review session records, action histories, and quality signals after launch.[1][2]
Production sessions and escalations feed an improvement process involving OpenAI Codex. OpenAI says Codex, using a Presence plugin, can investigate observed problems and propose updates. Customer teams compare a proposed version with the production version, test it, and approve a controlled rollout. The public materials describe Codex as proposing changes, not independently placing unreviewed changes into production.[1]
Security and privacy terms are not uniform across all Presence customers. OpenAI says each deployment separately defines accessible data, logging, masking of sensitive information, retention, storage location, and access controls. The approved architecture, security documents, and contract for a customer deployment are therefore the controlling sources. The general product documentation does not promise one fixed data-residency option, retention period, service commitment, or security configuration for every customer.[2]
Voice, chat, models, and APIs
During limited general availability, Presence supports conversational workflows through voice or chat. A deployment can combine real-time interaction with deeper reasoning or tool execution, but OpenAI does not publicly name a fixed model stack for the product. Model selection is validated for each workflow and may change over time.[1][2]
Presence is related to the OpenAI API through its use of OpenAI models and its ability to connect with business systems through APIs and tools, but it is sold and operated separately from self-serve API access. OpenAI said at launch that it would continue giving voice customers access to frontier models through the API. The OpenAI Realtime API remains the developer platform for building custom low-latency voice applications.[1]
GPT-Live is also a separate offering in the available documentation. OpenAI introduced GPT-Live on July 8, 2026, as the voice-model family behind a new ChatGPT Voice experience and said API access would follow. Presence launched two weeks later as a managed deployment platform. OpenAI's Presence documentation does not state that every Presence voice agent uses GPT-Live, nor does it bind Presence to a specific GPT-Live version. Any model relationship must be confirmed for the individual deployment.[2][4]
Launch uses and reported results
OpenAI's own English-language phone support line was the first named production example. The company said its Presence agent could handle open-ended requests, verify callers, use account context, and take approved actions. OpenAI reported that the system resolved 75 percent of inbound issues without human assistance and that its Codex-based improvement process reduced handoffs by 15 percentage points over 10 days. These figures were reported by OpenAI using its internal support benchmarks.[1]
The phone service also illustrates that Presence capabilities depend on configuration. OpenAI's help page says the support agent can answer routine product and account questions, but cannot submit reports, initiate account reviews, connect a caller to a live agent, or guarantee follow-up. Calls may be retained for service improvement, safety, and abuse prevention, and audio or transcripts may be used for model training after steps to reduce personal information.[3]
At announcement, OpenAI named three enterprise design or testing relationships. Their stated status was exploratory rather than a claim of broad production deployment:[1]
| Organization | Announced work | Status stated by OpenAI |
|---|---|---|
| BBVA Mexico | Voice support for everyday banking needs | Exploring as a design partner |
| SoftBank | Natural Japanese-language customer conversations | Testing |
| IAG | Support during high-demand events such as severe weather | Exploring |
Independent launch coverage also described Presence as a dedicated enterprise platform for building, deploying, monitoring, and revising voice and chat agents, and noted its Forward Deployed Engineer-led availability model.[6]
Access and limitations
Presence is not available through a public signup or self-serve console. Prospective customers must contact their OpenAI account team. Eligibility depends on workflow fit, implementation readiness, delivery capacity, and technical scoping. Pricing and implementation scope are negotiated for each customer.[1][2]
The limited general availability label also limits broad comparisons. OpenAI has not published standard prices, a universal feature matrix, fixed capacity, one list of supported integrations, or service-level commitments for all Presence deployments. Models, channels, authentication, human handoff, data treatment, and operational controls can differ between customers. OpenAI's published launch metrics cover its own phone-support deployment and should not be treated as guaranteed results for other workflows.[1][2]
References
- Introducing OpenAI Presence, OpenAI, July 22, 2026. ↩
- OpenAI Presence, OpenAI Help Center, updated July 2026. ↩
- AI phone support, OpenAI Help Center, updated July 2026. ↩
- Introducing GPT-Live, OpenAI, July 8, 2026. ↩
- ChatGPT Workspace Agents for Enterprise and Business, OpenAI Help Center, July 2026. ↩
- OpenAI's latest service wants to get your company build and get better integrated with AI agents, TechRadar, July 23, 2026. ↩
- OpenAI launches the OpenAI Deployment Company to help businesses build around intelligence, OpenAI, May 11, 2026. ↩
- Exclusive: OpenAI deployment arm to acquire Northslope, Axios, July 8, 2026. ↩
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation. Every suggestion is reviewed for sourcing before it goes live.