OpenAI Codex Remote Goes GA for All Paid Plans; Unsloth Bridges to vLLM and Ollama

TL;DR: OpenAI Codex Remote is now GA for all paid plans with QR-paired phone control; Unsloth v2026.6.7 adds a bridge to vLLM/Ollama so you can fine-tune locally and route inference out.

OpenAI Codex Remote — GA for all paid tiers

Codex Remote reached general availability on June 25 for every paid ChatGPT plan (Plus, Pro, Business, Enterprise, Education). The feature lets you start, review, and approve long-running Codex coding sessions from the ChatGPT mobile app while the agent runs on your Mac or Windows host.

The architecture change is the key detail: the old remote-shell approach required opening your dev machine to inbound connections. The new release replaces it with an authenticated QR-relay — each phone pairs one-to-one with one host, and your machine stays inaccessible to the public internet.

Why it matters: For Claude Code users, Codex Remote closes the mobile-control parity gap. Phone-based steering of AI coding agents is now table stakes across both platforms — useful context if you're evaluating a dual-tool workflow or pitching AI tooling to a team.

Unsloth + external inference backends

Unsloth v2026.6.7 added a bridge to external inference backends: after fine-tuning locally in Unsloth, you can now route inference to a vLLM server, Ollama, or llama-server without re-loading the model weights. This closes the local fine-tune to remote serve gap for teams that want GPU-accelerated training but hosted inference.

Sources: Work with Codex from anywhere — OpenAI · Codex Remote Connections — OpenAI Docs · Unsloth Changelog