OpenAI opened its Agents API to all developers in public beta on September 10, offering the agent runtime behind Codex as a managed service. The platform offloads multi-agent orchestration, context compaction, programmatic tool execution, and durable session state to OpenAI's managed infrastructure.
According to OpenAI’s documentation, the service currently supports data residency only in the United States and does not support Zero Data Retention (ZDR). Choosing a self-hosted sandbox does not change that limitation: OpenAI still manages the agent’s session state.
Billing Components
While OpenAI charges no separate platform fee for the Agents API itself, running an agent incurs charges across several billing categories:
- Model tokens: Model inference is billed at standard API rates.
- Tool calls: OpenAI tools use their standard published rates.
- Hosted execution: OpenAI-hosted container sandboxes bill at standard container rates.
Workspace Storage and Network Access
Developers can run code using self-hosted runners, partner infrastructure, or OpenAI-hosted environments. The hosted option supplies an isolated Linux workspace equipped with Python, Node.js, and command-line tools. Files placed in the session workspace persist across conversation turns throughout the container's operational lifetime, but sandbox retention is temporary: if both task activity and connection keep-alives cease for one hour, OpenAI may delete the sandbox.
To preserve run results beyond container teardown, files written to /workspace/outputs become immutable, downloadable artifacts upon turn completion and stay accessible after the sandbox expires.
Outbound internet access is enabled by default unless governed by a template policy. Teams can alternatively disable network access entirely or configure restricted exact-host filtering.