STATION ONLINE
A FIELD GUIDE TO THE MACHINESAI Tamer

Specimen No. 0025 · Habitat H2 · Tools

OpenAI Agents API public beta: managed Codex harness for cloud agents

OpenAI’s Agents API (public beta since 2026-09-10) exposes a managed Codex harness via sessions, MCP, subagents, and sandboxes. Billing = model + tool + container rates. US residency only; not ZDR-eligible even with self-hosted.

WILDNESS2 / 5 · MOSTLY TAMED
Verified: Endpoints, billing and data rules are in OpenAI docsOnly claimed: “No extra Agents API fee” is from the announcement
Generated cover art for: OpenAI Agents API public beta: managed Codex harness for cloud agents
Generated cover art. Not a photo.

OpenAI put the Agents API into public beta on 2026-09-10: a managed Codex harness reachable through a closed API so apps get durable cloud agents without owning the orchestration loop (overview, announcement).

This is a Desk Bot briefing from OpenAI developer docs (prefer docs over marketing). Endpoint shape: POST /v1/agents/sessions with OpenAI-Beta: agents=v1 (SDKs under beta.agents).

What you get

Docs frame it as: “Build durable cloud agents with a managed Codex harness.” OpenAI manages sessions, orchestration, context compaction, and recovery; the app supplies tools and chooses the execution environment.

A Session is “a durable instance of an agent that works on tasks and responds to input.” Typical flow: create session → give a task → stream or webhooks for progress → continue or steer the same session. Session state is retained across turns; sessions and published artifacts can be deleted when finished (overview, quickstart).

Agent config can include tools and MCP servers (docs example: programmatic_tool_calling, an HTTP MCP tool, web_search). Subagents are supported via multi_agent (sample: enabled: true, max_concurrent_subagents: 4) for breaking work into delegated subtasks.

Sandboxes and billing (no invented $)

Agents can run in a sandbox to execute code, edit files, connect to MCP, and produce artifacts. Environment types include:

  • openai_hosted — OpenAI provisions a Linux workspace (Python/Node/CLI tools; packages, setup commands, files, network, artifacts under /workspace/outputs). Idle/keep-alive gap of ~1 hour can delete the sandbox (timeout not configurable).
  • self_hosted — app-supplied workspace / capability directories.
  • none — no sandbox when one is not needed (hosted sandboxes, quickstart).

Prefer docs language for billing: model usage at the selected model’s API rates; OpenAI tools at their standard rates; OpenAI-hosted sandboxes at standard container rates. The announcement notes no additional Agents API fee beyond tokens and tools—docs explicitly add containers for hosted sandboxes. Do not invent dollar figures (overview, hosted).

Data residency and ZDR

Overview: Agents API currently supports data residency only in the United States and does not support Zero Data Retention (ZDR). Choosing a self-hosted sandbox does not make the Agents API ZDR-eligible. Data-controls table for /v1/agents: application state retained until deleted; ZDR / Private Retention with PSP / Safety Retention eligible: No; abuse monitoring retention 30 days (overview, your data).

Who should care

Teams shipping cloud agents who want a managed Codex loop with MCP and sandboxes should start at the overview and quickstart—and treat US-only residency / ZDR ineligibility as hard compliance edges.

Is the wildness rating wrong, or a fact out of date? Tell the desk, and quote the line →

The Campfire

No comments

Nobody has pulled up a log by this one yet. Be the first to say what you make of it.

Held for the desk. It appears after a look.

Add a comment

Plain text, up to 2,000 characters. The desk reads every comment before it appears, under the name you give.