Codex Web GPT v4.0.0

V4 turns ChatGPT Web into a continuous Codex task runtime. Multiple Codex messages can now stay in the same task-bound Temporary Chat, and a full context rolls into a clean chat without losing the agent that actually did the work.

The big one: native context compaction

When Codex reaches its context boundary, it sends a control request to the exact retained Web agent through the gRPC-backed MCP tunnel. That agent produces the checkpoint summary, the bridge returns it to Codex, and only then retires the old Temporary Chat and starts a clean compaction epoch from that summary.

The browser conversation no longer has to grow forever, and compaction no longer asks an unrelated fresh chat to reconstruct the task from a giant replay.
If the retained private chat was manually closed or lost after a restart, V4 safely falls back to packaging the canonical Codex history into one fresh, tool-free compaction chat.

Also in V4

• Continuous Web conversations. Sequential Codex messages reuse the same Temporary Chat until compaction. Every message gets a fresh turn-bound MCP capability, while all of its tool rounds stay inside one Web response.
• Bigger Context is substantially more stable. The experimental mode keeps small tasks on one message and adapts large contexts across two or three verified messages; the final part starts the real task without a redundant request. It can expose up to 3× the normal context and compaction limits. Extra requests can still increase the chance of rate limits or temporary cooldowns.
• More resilient streaming and reconnects. Response tracking now survives React DOM remounts, hidden tabs keep an operational viewport, active MCP calls count as proven progress, reconnects rejoin the existing browser turn, and already-streamed text is preserved.
• ChatGPT UI changes handled. Connector turns enable Personalized only when required, current effort sliders and model rows are detected, and Japanese cooldown dialogs are recognized.
• Launcher and platform fixes. Session-refresh/login races are serialized, exported diagnostics stay privacy-safe, Linux AppImages ship a verified compatible libnotify, and their FUSE/extraction lifecycle is bounded.

Addresses #146, #149, #170, #179, and #181. Also includes #171, #172, and #176.

Install

macOS or Linux:

curl -fsSL https://github.com/miuuyy/codex-chatgpt-web/releases/latest/download/install-launcher.sh | sh

Windows PowerShell:

irm https://github.com/miuuyy/codex-chatgpt-web/releases/latest/download/install-launcher.ps1 | iex