OpenAI announced the public beta release of the Agents API, which brings the harness and managed infrastructure behind Codex to developers via a single, flexible API. The goal is to let developers create production-ready, long-running agents with one API call by specifying the task, model, tools and environment.
What the Agents API provides
- A harness hosted and maintained by OpenAI: the same core logic that manages context, tool usage and subagent coordination used to scale Codex and ChatGPT for Work.
- Choice of compute environment: developers can run agents in an OpenAI-managed sandbox, on their own infrastructure, or with a partner-managed sandbox.
- OpenAI-hosted sandbox: for quick starts and efficient scaling, OpenAI provides sandbox environments that run code, handle files and produce artifacts. These sandboxes are provisioned and managed by OpenAI and can be configured with files, packages, skills and plugins required by the agent.
Ecosystem partner integrations
The Agents API offers first-class integrations with a set of ecosystem providers to cover a range of compute, storage and deployment needs. Named partners include Blaxel, Cloudflare, Daytona, DigitalOcean, E2B, Modal, Oracle, Runloop and Vercel.
Technical capabilities highlighted
The announcement outlines several features designed to make long-running agents practical:
- Context management and automatic compaction: the API helps agents persist relevant information across longer sessions by automatically compacting earlier context as a session approaches its context limit, allowing workflows to span multiple context windows without custom compaction logic.
- Tool search: loads relevant tool definitions on demand to reduce token usage and cost while preserving the model’s cache.
- Programmatic tool calling: enables agents to run calls in parallel, chain operations, and filter or combine results in code, which makes it easier to process large volumes of data. The Agents API supports MCP, custom functions and built-in tools such as web search.
- Multi-agent support: complex tasks can be split into independent pieces and delegated to subagents that work in parallel; each subagent keeps its own context while the main agent coordinates and aggregates results.
Open-source harness and versioned access
The Agents API is powered by the open-source Codex harness, giving developers visibility into the core coordination logic for model calls, tools and context. OpenAI operates and maintains the managed harness, and the public codebase is available for inspection.
Pricing and availability
The Agents API is available in public beta to all developers. OpenAI states there are no additional Agents API fees; developers pay for the tokens and tools their agents consume, according to OpenAI’s pricing information.
Practical implications
By exposing the Codex harness and managed sandboxes, OpenAI aims to let developers focus on the unique tools, knowledge and workflows that differentiate their agents while OpenAI handles the underlying infrastructure and harness improvements. During the public beta OpenAI will iterate quickly based on developer feedback and invites reports on what works, where friction appears, and what is needed to run agents in production.
