OpenAI released its Agents API in public beta on Thursday. The release gives developers the same infrastructure that powers the company’s Codex coding agent and ChatGPT for Work, accessible through a single API call. The move formed part of a larger set of product launches that day, which also included a Data agent for enterprise users and developer access to the GPT-Live-1 voice model.
How the Agents API Works
Developers create cloud-based agents by defining a task, model, tools, and compute environment. OpenAI handles the orchestration harness that keeps everything running. Agents can operate inside OpenAI-hosted sandboxes, on a company’s own infrastructure, or through partner integrations that include Cloudflare, DigitalOcean, Oracle, Vercel, and additional providers.
The API adds several technical capabilities. Automatic context compaction manages sessions that last for hours. Tool search lowers token use. Programmatic parallel tool calling lets multiple tools run at once. Multi-agent support hands subtasks to subagents that execute concurrently.
Jack Weissenberger, CTO of Ciridae, reported that the API lifted the company’s evaluation score from 0.71 to 0.85 and delivered a fourfold drop in latency for subagent workflows. Pricing stays limited to standard token and tool usage charges. The harness itself draws from OpenAI’s open-source Codex codebase.
Data Agent Inside ChatGPT Work
OpenAI also introduced a Data agent within ChatGPT Work. The agent connects directly to enterprise data sources such as Amazon Redshift, Google BigQuery, Snowflake, Databricks, and MongoDB. It pulls context from data warehouses, Slack, file storage, and business intelligence dashboards so users can ask questions without first consolidating the underlying data.
Arpan Shah, general manager of OpenAI’s enterprise technology vertical, described the change inside the company: “If you asked me a year ago what percentage of our data questions were getting answered end to end without needing anyone intervening, it was a very small fraction of the company. Today, everyone at OpenAI can self-serve any data questions they have.”
GPT-Live-1 Voice Model for Developers
The same day, OpenAI made GPT-Live-1, its full-duplex voice model, available through the API. Yelp integrated the model into its Yelp Host restaurant call-handling tool and its Hatch communications platform. Those two products have handled more than a million restaurant calls since they launched in October 2025.
Availability and Next Steps
The Agents API opened to all developers on Thursday. OpenAI plans to gather feedback and refine the service before moving it to general availability. The managed approach removes the need for developers to maintain their own harness, manage context, or handle multi-agent orchestration—capabilities the company refined while running Codex at scale. That positions the API as a direct alternative to other agent frameworks.