Agents API is an Automation tool. Managed API for building cloud agents with orchestration, long-running sessions, and tool use. Key features include Codex Harness Management, Automatic Context Compaction, and Multi-Agent Orchestration. Best for software developers and engineers, data scientists and analysts and business consultants.
About Agents API
Key Features
<strong>Codex Harness Management.</strong> OpenAI manages the underlying agent infrastructure so you can focus on what your agents do, not how they run.
<strong>Automatic Context Compaction.</strong> The API automatically compacts earlier context as sessions near their limit, eliminating the need to write your own compaction logic.
<strong>Multi-Agent Orchestration.</strong> Split complex tasks into independent pieces with subagents that keep their own context while the main agent coordinates and combines results.
<strong>Programmatic Tool Calling.</strong> Agents run tool calls in parallel and chain operations, filtering or combining results in code so only relevant data returns into context.
<strong>Long-Running Sessions.</strong> Durable sessions work on tasks for days with environments where agents can access files, run code, and save intermediate results.
<strong>Flexible Deployment Options.</strong> Deploy agents in OpenAI-managed sandboxes, on your own infrastructure, or with sandbox partners like Cloudflare, DigitalOcean, and Oracle.
Frequently Asked Questions
The Agents API is a managed service from OpenAI that gives developers access to the Codex harness through a simple API. It handles sessions, orchestration, context compaction, and recovery while you provide tools and choose your execution environment. You can create production-ready agents in a single API call.
The Agents API itself has no extra fee. You pay only for model API calls at standard token rates, plus OpenAI tools at their standard rates and OpenAI-hosted sandboxes at standard container rates. Model usage is billed at the selected model's API rates.
The Agents API is built for software developers and engineers who want to build long-running, multi-functional agents without managing the underlying infrastructure. It's ideal for teams building incident response agents, Slack bots, data analysts, or any workflow that needs durable sessions and tool orchestration.
Agents built with the Agents API can execute code, edit files, connect to MCP servers, produce artifacts, and coordinate subagents. They can handle complex tasks like investigating alerts, answering warehouse questions with SQL, or comparing release notes across multiple sources.
