OpenAI Agents API adds computer use
OpenAI documents computer use as an additional Agents API capability, extending the managed Codex harness beyond tools and multi-agent workflows.
OpenAI documents computer use as an additional Agents API capability, extending the managed Codex harness beyond tools and multi-agent workflows.
The Agents API provided a managed Codex harness with context, tools and subagents.
OpenAI additionally documents computer use as a supported Agents API capability.
OpenAI added computer use to the Agents API in its DevDay 2026 recap. The API provides developers with an OpenAI-operated Codex harness that manages context, uses tools and can coordinate multiple agents. OpenAI now also describes computer interaction as part of that managed agent runtime.
For AgentenCode, this is a meaningful capability change because computer use can expand the action surface beyond explicitly integrated APIs or tools. Depending on the configured environment and permissions, an agent may also interact with graphical software interfaces.
That broader action surface makes governance more important. Computer interactions can change external state, so permission boundaries, human approval, logging and safe execution environments become more consequential. AgentenCode records the capability only within the documented Agents API scope.
Availability and authorization still need to be separated. The presence of computer use does not imply that every application can access arbitrary interfaces without additional configuration or approval. Product plan, runtime setup and specific permissions remain separate evidence questions.
Primary source
What this update is based on
AgentenCode publishes product changes only after source-first review. The original provider source remains authoritative for current details.
Continue checking
Full context in the Agent Profile.
The profile, trust evidence and history separate documented properties from Unknown.