OpenAI makes the Codex harness available through the Agents API

OpenAI moved the Agents API into public beta today. The hosted service runs the same harness and infrastructure behind Codex, including session orchestration, context compaction across long sessions, tool search, and multi-agent delegation.

Developers can run the work in an OpenAI-managed sandbox, on their own infrastructure, or in a partner environment from Cloudflare, DigitalOcean, Modal, and six other providers. OpenAI charges for the tokens and tools an agent uses, with no additional Agents API fee.

OpenAI's announcement includes several customer-reported gains. SafetyKit's Bhavyansh Sabharwal says migrating a case-review workflow cut cost per case by 60 percent. Hypha's Serhii Shchoholiev reports that separating the harness from the sandbox cut failed agent responses by 86 percent. Ciridae's Jack Weissenberger says the API moved an evaluation score from 0.71 to 0.85 and cut subagent latency fourfold.

Those figures come from testimonials selected by OpenAI, not independent benchmarks. None of the seven customers quoted in the announcement reports a downside.

On the Hacker News thread discussing the launch, programmer Simon Willison tested the sandbox's network controls directly. He asked an agent to add a spoofed DNS entry and bypass the allowed-domains list with a forged Host header. Willison reported that the attempt was "rejected," with the sandbox returning "Domain forbidden."

Why it matters: Agent harnesses have mostly lived inside each vendor's own CLI. The Agents API turns OpenAI's harness into a managed service. Teams choosing between building an orchestration layer and buying one now have specific, vendor-reported cost and reliability comparisons to test.