OpenAI ships Codex CLI 0.134.0

OpenAI released Codex CLI version 0.134.0 on May 26, 2026, continuing a rapid cadence of feature and stability updates to its terminal agent tooling.

The release is published in the official Codex changelog and shows the standard install command: npm install -g @openai/codex@0.134.0.

Among the headline additions is a local conversation search that scans a session’s history, supports case-insensitive matches, and shows result previews to help developers find prior context quickly.

The CLI also consolidates profile handling: the new build makes --profile the primary selector for CLI actions, TUI permission flows, and sandbox environments, and it includes migration guidance that rejects legacy profile configs.

On the MCP (managed connector/proxy) side, Codex 0.134.0 adds per-server environment targeting and OAuth options for streamable HTTP servers, and it surfaces profile-migration guidance under the new --profile handling.

A notable runtime improvement lets MCP tools that advertise a readOnlyHint run in parallel, a change designed to reduce blocking and improve throughput for tasks that only read external state.

The changelog also lists multiple small but meaningful fixes: TUI keymap coverage improvements, clearer permission-profile wiring, and Windows sandbox/PTTY teardown fixes that address rendering and teardown corruption.

These updates are part of a pattern across vendors: terminal-first agent tooling keeps gaining production features such as per-environment targeting, stronger permission profiles, and richer memory and history handling. OpenAI’s CLI worklines up with similar moves by other vendors to make agents safer and more reliable in developer workflows.

For developer teams, the practical benefits are straightforward: searchable local conversation history reduces time lost hunting previous prompts, unified profile flags cut configuration friction during CI and developer-machine testing, and parallel read-only MCP calls can speed composite agent runs.

The release notes also point to ongoing attention to telemetry, diagnostics, and test stability — changes that reduce flakiness in CI and help teams debug agent runs in production. Those back-end improvements are less visible day-to-day but important for long-lived agent deployments.

Security and supply-chain hygiene remain in the background of these updates: OpenAI has pushed several client and signing fixes in recent months, and administrators should follow vendor guidance when upgrading CLI, desktop, and sandboxed components.

If you run Codex in a team or production setting, the immediate steps are simple: upgrade to 0.134.0, test profile migration in an isolated workspace, and validate MCP environment targeting in staging before rolling changes to live agents. The changelog and docs include links and migration notes to guide that process.