OpenAI has launched the Agents API, a new API that lets developers build and run their own cloud-based AI agents. The service is based on the same agent management system the company uses in Codex.
OpenAI has launched the Agents API, a new API that lets developers build and run their own cloud-based AI agents. The service is based on the same agent management system the company uses in Codex.
This is reported by OpenAI.
Now, instead of having to independently collect "memory" for the agent, connect to tools, create an environment for running code, and manage multiple models, you can delegate much of this work to the Agents API.
According to OpenAI, a worker agent can be launched with a single API call: it is enough to specify the task, model, available tools, and the environment in which it will work. After that, the agent can perform long tasks, use external tools, work with files, and run code.
For example, if a service suddenly starts to experience a spike in errors, an agent can be tasked with investigating what happened. They can review recent changes, logs, and dependencies, then draw conclusions and suggest how to fix the problem.
The Agents API also supports subagents. A large task can be divided among multiple AI assistants that will work in parallel. In the same example with a failure, one agent can check for updates, another for errors, and a third for external dependencies. The master agent coordinates them and collects the results into a single response. In addition, each subagent has its own context, so it does not need to keep all the information about the task in its “head”.
For longer tasks, OpenAI has added automatic context management. When the work history approaches the limit of the context window, the Agents API compresses previous information but tries to preserve what the agent will still need. The Agents API can also select the right tools when they are needed. MCP, developer-specific functions, and OpenAI’s built-in tools, including web search, are supported.
You can run your agent on your own infrastructure, use environments from OpenAI partners — including Cloudflare, DigitalOcean, Vercel, Oracle, and others — or choose the new OpenAI-hosted sandbox. In the latter case, OpenAI itself creates an isolated environment in which the agent can run code, open and modify files, and save the results of its work. You can add your own files, packages, skills, and plugins to this sandbox.
The Agents API is now available to all developers in public beta. There is no separate fee for the API itself. The user pays only for the model tokens and tools used by the agent.
Previously, dev.ua wrote that OpenAI announced the creation of an "automated research intern" - an AI system that, under human guidance, can perform research tasks that would take a specialist several days to complete.
