AI
OpenAI documents Agents API for persistent sessions with managed Codex orchestration
OpenAI’s Agents API exposes a managed Codex harness that handles orchestration, context compaction and recovery while applications choose tools and execution environments. Developers can maintain sessions across tasks, stream progress and steer work in progress, potentially reducing infrastructure needed for persistent agents. The documentation specifies separate model, tool and hosted-container charges. It also establishes a significant deployment constraint: US-only data residency and no Zero Data Retention support, including with self-hosted sandboxes.
Editorial summary · Publisher article
What matters
- Sessions preserve state so applications can continue work without rebuilding conversation context.
- Supported capabilities include sandbox execution, MCP connections, skills and parallel subagents.
- Applications can receive progress through streaming or webhooks.
- Self-hosting the sandbox does not change the API’s Zero Data Retention restriction.
Context & caveats
- The documentation describes supported capabilities; it does not establish reliability or performance in production.
Why this made the edition
Primary documentation supplies concrete architecture, integration, billing and data-retention details relevant to developers.
Source material