OpenAI Presence Explained: Features, Pricing, and Limits

OpenAI Presence is a managed enterprise platform for deploying AI agents into high-volume, repeatable workflows. Announced on July 22, 2026, it combines OpenAI models with business-system connections, policies, evaluations, monitoring, and human escalation. Early use cases focus on voice and chat for customer support, outbound sales, and internal service workflows.

Presence is not a self-serve agent builder or a new foundation model. It is a limited-availability deployment product shaped with OpenAI engineers and selected implementation partners. That distinction matters for buyers comparing it with ChatGPT Workspace Agents, the OpenAI API, or a conventional contact-center bot.

What Is OpenAI Presence?

OpenAI describes Presence as a platform for building, launching, operating, and improving governed AI agents. A deployment starts with a defined job, such as resolving billing questions, supporting an insurance claim, or handling an employee IT request. The agent receives the knowledge, tools, and permissions needed for that job rather than broad access to every company system.

The product is intended to cover more than the conversation layer. Its documented components include standard operating procedures, policies, guardrails, approved actions, simulations, evaluation tools, production monitoring, and a process for testing changes before rollout.

That makes Presence closer to a managed operating system for a specific agent workflow than a chatbot subscription. The model can reason and use tools, but the deployment also has to answer operational questions: What can the agent change? Which requests require approval? When should a person take over? What evidence is recorded? How is a weak version rolled back?

How OpenAI Presence Works

OpenAI's Help Center outlines a six-stage managed deployment process. The exact sequence and implementation can vary, but the public framework is useful for understanding what a buyer is purchasing.

1. Define The Workflow And Success Criteria

The customer and deployment team select a bounded workflow and define measurable outcomes. A useful target is usually frequent, structured enough to evaluate, and supported by clear policies. “Help with every customer problem” is difficult to test. “Resolve eligible billing disputes under a documented threshold and escalate exceptions” is much more concrete.

2. Connect Knowledge, Systems, And Permissions

Presence can use approved knowledge and connect to business systems through tools and APIs. Permissions are meant to be scoped to the workflow. A support agent may need to read an account record and issue a permitted adjustment, for example, without receiving unrestricted access to the full customer database.

This is where ordinary integration work remains important. Identity, authentication, data quality, API reliability, audit records, and failure handling do not disappear because an AI model is involved.

3. Encode Policies And Escalation Rules

Teams define approved instructions, business policies, actions, approval steps, and situations that require human judgment. The agent should know both how to complete the normal case and when to stop.

That approach matches the practical controls in SD's guide to AI agent security: useful autonomy depends on least privilege, explicit boundaries, logging, and meaningful human intervention.

4. Test Before Production

OpenAI says Presence deployments use simulations, evaluations, and acceptance testing across common requests, edge cases, policy compliance, tool use, and escalation behavior. A model answering correctly in a demo is not enough. The combined system must also take the right action, avoid forbidden actions, and recover safely from ambiguous or broken inputs.

5. Roll Out Gradually

The documented process includes staged production rollout and monitoring rather than an immediate switch to full automation. This gives teams a chance to compare outcomes, review sessions, and catch workflow gaps while exposure remains limited.

6. Improve From Production Evidence

Session records, action histories, escalations, and quality signals can show where the agent fails. OpenAI says Codex can investigate those signals and suggest updates, which teams then test against the production version before approving a controlled release. The important control is approval: an improvement proposal is not supposed to become a live behavior change automatically.

OpenAI Presence Features That Matter

Voice And Chat Workflows

During limited general availability, Presence supports conversational workflows through voice or chat. OpenAI lists customer support, outbound sales, and higher-risk internal workflows as current examples. Specific channels, contact-center integrations, authentication, routing, and handoff designs are determined during deployment.

Approved Business Actions

A Presence agent can retrieve information, update systems, and complete actions when those actions have been approved and connected. This is more consequential than answering from a knowledge base. It also raises the standard for permissions, testing, and auditability.

Evaluations, Guardrails, And Rollbacks

OpenAI documents simulations, graders, permissions, approval steps, production monitoring, human escalation, controlled releases, and rollback processes. These are essential parts of the product proposition because a production agent can fail through a bad answer, an incorrect tool call, a stale policy, or a technically valid action applied to the wrong case.

Managed Deployment Support

OpenAI Forward Deployed Engineers, selected systems integrators, or both work with the customer. This can help enterprises that have a valuable workflow but lack the internal agent engineering, evaluation, or deployment capacity to operate it safely. It also means Presence is not designed for buyers seeking an instant signup and a fixed feature list.

OpenAI Presence Performance: What The Evidence Shows

OpenAI reports that Presence powers its English-language phone support line at 1-888-GPT-0090. According to the company, the system handles open-ended requests, verifies callers, uses account context, and takes approved actions. OpenAI says it now resolves 75% of inbound issues without a human and that an improvement loop reduced handoffs by 15 percentage points in 10 days.

Those figures are relevant but should be treated as vendor-reported results from one deployment. OpenAI has not published the underlying case mix, evaluation rubric, error distribution, customer-satisfaction data, or an independent audit on the launch page. Results should not be generalized to another company, language, workflow, or risk level without a representative pilot.

OpenAI also names BBVA, SoftBank, and IAG as design partners exploring or testing voice-support scenarios. Their statements demonstrate enterprise interest, not proof of generally available performance across industries.

OpenAI Presence Pricing And Availability

There is no public OpenAI Presence price list as of August 1, 2026. Pricing, implementation scope, models, channels, capacity, data handling, and service commitments are defined for each deployment.

Presence is available through limited general availability for eligible enterprise customers. Access depends on workflow fit, implementation readiness, and delivery capacity. Organizations must contact their OpenAI account team; there is no self-serve checkout or public API endpoint specifically for creating a Presence deployment.

Buyers should evaluate total cost rather than asking only for a model rate. Likely cost categories include implementation, system integration, evaluation development, contact-center or channel work, ongoing monitoring, human escalation, change management, and usage. OpenAI has not publicly disclosed how these categories are packaged, so any detailed price estimate would be speculation.

OpenAI Presence vs Workspace Agents And API Agents

OpenAI Presence and ChatGPT Workspace Agents both automate repeatable work, but they serve different deployment needs.

• OpenAI Presence is a managed product for high-volume production workflows that need deep integrations, testing, guardrails, monitoring, escalation, and deployment support.
• ChatGPT Workspace Agents are created and managed inside supported ChatGPT and Slack workspaces. Teams can select models, connect approved apps, share agents, schedule runs, or use API triggers.
• API-built agents give developers the most direct control over architecture, interfaces, tools, observability, and release processes, while leaving more engineering and operational responsibility with the customer.

For individual knowledge work and long multi-step deliverables, SD's ChatGPT Work review covers a different OpenAI product category. Presence is aimed at a company deploying a governed workflow for customers or employees, not a user asking an agent to complete one project.

The practical choice is based on operating model. Use a workspace agent when the workflow fits existing ChatGPT or Slack experiences and internal teams can configure it. Build with APIs when custom product control is the priority. Consider Presence when the workflow is valuable enough to justify managed implementation and ongoing production governance.

Who Should Consider OpenAI Presence?

Presence appears best suited to organizations with all of the following:

• a high-volume, repeatable workflow with measurable outcomes;
• documented policies and clear escalation conditions;
• systems that can expose safe, scoped actions;
• enough historical cases to build representative evaluations;
• operational owners who can review quality after launch;
• security, privacy, legal, and procurement capacity for a managed deployment.

It is a weaker fit for early brainstorming, low-volume ad hoc work, teams that need transparent public pricing, or buyers still trying to identify a stable use case. It may also be excessive when a read-only workspace agent or a narrow API application can meet the same need with lower integration cost.

Before procurement, ask for a pilot design that defines the baseline, target metrics, unacceptable failure modes, human-review plan, rollback criteria, and full operating cost. A successful demo should not be the acceptance test.

Conclusion

OpenAI Presence is a serious enterprise-agent offering because it focuses on the operational layer around the model: scoped access, policies, evaluations, monitoring, human escalation, and controlled improvements. That design matches what high-volume production workflows require, but the launch evidence is still mostly supplied by OpenAI, availability is limited, and pricing is private.

The strongest buying case is a narrow, valuable workflow with clear rules, safe system actions, measurable outcomes, and an accountable human owner. Enterprises should judge Presence through a representative pilot and total operating cost—not launch metrics alone. Teams that need self-serve automation or maximum architectural control may be better served by Workspace Agents or API-built systems.