Codex (OpenAI): Definition
Codex is OpenAI's code-execution runtime, the engine that lets agents run code, handle multi-step execution, and interact with external systems inside Workspace Agents and related products.
Longer definition
Codex originated as OpenAI's code-generation model in 2021, later becoming the underlying runtime for advanced agent behaviors. In 2026, Codex powers the execution layer of Workspace Agents: the ability for an agent to run Python code in a sandbox, chain tool calls with code logic, and handle multi-step workflows. When OpenAI says a product is "Codex-powered," they mean it uses this runtime for execution, not just model inference.
In practice
For most users, Codex is invisible. It's the engine behind the agent, not something you interact with directly. It matters when you're trying to understand what your agent can actually execute (code in a sandbox, not arbitrary code on your systems) and why certain execution latencies exist.
Common mistakes
- Thinking Codex means the agent can run code on your servers, when it runs in OpenAI's sandbox, not your infrastructure
- Confusing Codex-the-runtime (2026) with Codex-the-original-model (2021), which are related but not the same
- Expecting Codex-powered agents to handle tasks requiring persistent external servers, since sandbox execution is ephemeral
See also
Want this shipped for your team?
20 minutes on video. We'll tell you if it's a fit and what it would take.