What's the difference between the OpenAI Agents SDK and the Codex SDK?

The OpenAI Agents SDK is a general framework, a few primitives, agents, handoffs, and guardrails, for building any agent you design; the Codex SDK is narrower, a library that only starts, resumes, and streams threads on OpenAI’s own Codex coding agent.

The Agents SDK gives you pieces to assemble: which tools an agent carries, which specialist it hands a task to, what its guardrails check going in and out. It defaults to OpenAI’s Responses API and reaches other providers only through adapter packages still in beta; OpenAI’s own docs call it “a production-ready upgrade of our previous experimentation for agents, Swarm.” The Codex SDK gives you no pieces at all. It drives one fixed agent, the harness behind the Codex CLI, app, and IDE extension: startThread() opens a run, and a past thread resumes or streams by its id, enough to trigger Codex from a CI pipeline, but what Codex does isn’t yours to configure.

Reach for the Agents SDK to design an agent’s own logic. Reach for the Codex SDK only to run OpenAI’s own Codex on your code as is. Underneath either one, tool calling works the same way: a structured call naming a tool and its arguments.

sources

keep reading

More on this.

Do OpenAI Agents SDK guardrails run on every agent after a handoff?No. Input guardrails only check the first agent and output guardrails only the last, so an agent picked up mid-handoff runs with no guardrail of its own.answer →Does the OpenAI Agents SDK support MCP tools?Yes. The SDK ships dedicated MCP server classes, MCPServerStdio, MCPServerStreamableHttp, and HostedMCPTool, that an agent uses like any other tool.answer →How do I catch a regression in an Agents SDK workflow?Grade production traces first to see what a change broke, then promote the cases that matter into a fixed dataset you rerun and compare after every change.answer →Everything on openai agents SDK evalsThe definition, and every question answered under openai agents SDK evals.concept →

Two ways to run Tessary.

Tessary is an open-source agent reliability platform. Cloud and self-hosted run the same workflow on the OpenTelemetry traces your agent already emits.

Tessary Cloud

We host it for you. Send your first trace with nothing to deploy and no model key.

what's includedper organization
traces
10,000 per calendar month
stored trace data
1 GB
retention
30 days
model credit
$10, one-time, for triage and root-cause analysis
credit card
not required
start for free

Self-hosted Tessary

Run the open-source code on your own infrastructure with one command. Add your own model key for triage and root-cause analysis.

Self-host Tessary for me by following https://github.com/tessaryai/tessary/blob/main/setup.md

docker compose -f oci://docker.io/tessaryai/tessary:compose up -d -y