OpenAI Expands Codex Tooling, GitHub Copilot App, LangChain Agents

OpenAI has broadened the reach of its Codex coding agent by embedding it into the ChatGPT mobile app. The new integration lets users initiate tasks, review outputs, approve commands, and steer execution from their phones while Codex continues to run on a laptop, Mac mini, or dedicated devbox. Remote SSH support is now generally available for managed remote environments, and the company has released hooks and programmatic access tokens to enable business‑level automation around the Codex loop. A technical write‑up on a Windows sandbox was also published, detailing how to balance utility with constrained machine access through local users, firewall rules, ACLs, write‑restricted tokens, DPAPI, and helper executables.

Meanwhile, GitHub unveiled a technical preview of the Copilot App, a desktop environment that supports parallel workstreams, repository and pull‑request lifecycle management, and model flexibility. Visual Studio Code added a new Agents window that supports multi‑agent, multi‑project workflows, browser and mobile support via vscode.dev/agents, BYOK improvements, and compressed terminal output. Other tools have also integrated Codex runtime: Nous/Hermes Agent routes OpenAI‑backed turns through Codex CLI/app‑server, while Kimi’s Web Bridge browser extension exposes web interaction to a range of coding agents, including Claude Code and Cursor.

LangChain has released a suite of agent lifecycle infrastructure, including LangSmith Engine, SmithDB, Sandboxes, Managed Deep Agents, LLM Gateway, Context Hub, and Deep Agents 0.6. SmithDB offers 12–15× faster access on key workloads and is built on Apache DataFusion and Vortex. CoreWeave launched Sandboxes for isolated execution in reinforcement learning, tool use, and evaluation workloads, testing destructive commands at scale. Open‑source and local development tooling has surfaced around agent debugging, with traces exposed to Codex/Claude Code for automated evaluation authoring.

Anthropic tightened restrictions on Claude Code usage for third‑party wrappers and high‑volume programmatic workflows, sparking backlash from developers who cited dramatic rate‑limit reductions and subscription cancellations. The company partially offset this by raising Claude Code weekly limits by 50% through July 13. In response, OpenAI offered two months of free Codex usage for enterprise customers who switch within the next 30 days and published a detailed Windows sandbox design to help users safely run coding agents with local filesystem and tool access.