vibe·code24/7
← all news
·4 minopenaicodexchatgptvoicedesktopenterpriseupdates

openai codex gets voice, multi-folder projects, and cloud sync in chatgpt desktop update

A July 2026 ChatGPT desktop update brings voice control into Work and Codex, lets local projects span multiple folders (with a primary root for Git and agent config discovery), adds a global switcher between ChatGPT and Codex, and syncs cloud Work sessions across devices. Separately, OpenAI launched Presence, an enterprise voice-and-chat agent product that resolves 75% of inbound issues without human escalation.

OpenAI pushed a set of ChatGPT desktop updates in July 2026 that are primarily about making Codex feel less like a separate product and more like part of the same workspace as ChatGPT. Voice, multi-folder project support, a unified switcher, and cross-device cloud sync are the headline changes.

voice in work and codex

ChatGPT Voice is now available inside Work and Codex in the desktop app. You can speak naturally, interrupt mid-sentence, and ask Voice to start or coordinate work using any tool or permission you've connected.

The practical implication: you can now launch an agent run, redirect a background task, or ask Codex to explain what it's currently doing — all without touching a keyboard. The interrupt-and-redirect capability is the part that makes this more than a dictation layer; it's a live control surface for running agents.

The integration uses the same connection model as other Voice-enabled tools — so if you've connected Gmail, Slack, or other workspace tools, Voice can coordinate with those through Codex without additional setup.

multi-folder local projects

Local projects in the desktop app can now include multiple related folders rather than a single root directory. The model designates a primary folder for new chats, Git operations, and automatic discovery of AGENTS.md, skills, and config.toml. Secondary folders remain available for file search, reading, and editing across the project.

This matters for the common repo setup where your monorepo root is separate from the tooling config, or where you're working across a frontend repo and a backend repo simultaneously. Previously you either had to pick one root and manually reference paths in the other, or symlink things together. Multi-folder projects solve this cleanly.

The AGENTS.mddiscovery behavior is worth noting specifically: only the primary folder is scanned for agent configuration files. So if you have AGENTS.md files scattered across secondary folders and expect them to be picked up automatically, they won't be — they need to live in or under the primary root.

unified chatgpt / codex switcher and cloud sync

The desktop app now has a global switcher that moves between ChatGPT conversations and Codex sessions from a single interface, alongside unified Recents that surfaces both chat and Work conversations in one list.

Cloud Work syncing means your in-progress Codex sessions are accessible from other devices without manual export or copy-paste. If you start a session on your desktop and need to hand off to someone else's machine, the session state travels with your account.

Projects support was also added to the desktop app in the same update — letting you organize Work sessions into named project groups the same way you can organize chat conversations.

openai presence: enterprise voice agent product

Separate from the desktop update, OpenAI launched Presence as a deployed product for enterprise customers. Presence pairs model reasoning with policies, guardrails, and escalation rules to handle voice and chat inbound volume — OpenAI says it resolves 75% of inbound issues without human assistance.

The product targets the same use cases as incumbent voice agent platforms (Five9, Verint, NICE) — customer support, internal helpdesk, appointment management — but built on GPT-5.6 reasoning rather than intent classification trees. The 75% resolution figure, if it holds in production, is significantly above the 40–60% baseline for intent-based IVR systems.

Presence isn't a Codex product — it's a separate enterprise deployment of OpenAI's model capabilities. But it signals where OpenAI is putting commercial weight: not just in the dev-tools market Codex competes in, but in the customer-facing enterprise software market where the deal sizes are larger and the competition is more entrenched.

stability fixes worth knowing

The July update also included warnings when running Ultra reasoning with high multi-agent concurrency (a combination that was previously allowed without notice even though it could saturate rate limits), fixes for Code Mode crashes on Intel macOS, Windows sandbox file deletion capabilities, prevention of terminal control sequence corruption, and authentication refresh for long-running sessions.

The Intel macOS crash fix is notable because it's been a persistent complaint — Codex's Code Mode has been unreliable on Intel Macs since launch, and this is the first fix that directly addresses the crash rather than a workaround.