Codex Web GPT v5.0.7

This is the last planned V5 patch before the major 6.0 update. It brings together the tested fixes below while the broader investigations and larger changes continue toward 6.0.

V5 adds Zero Risk Mode: the full Codex MCP harness still runs through ChatGPT, but the launcher never reads or changes the ChatGPT page and never sends a prompt for you.

V5 highlights

• Zero Risk Mode. The launcher prepares and copies the complete Codex prompt; you paste it into ChatGPT, add any images yourself, select the Codex Zero Risk connector plus the model and effort you want, send it, and confirm Sent. Codex tools, patches, terminal calls, structured output, streaming, and native progress stay available through the same turn-bound harness without browser automation.
• A separate Zero Risk MCP setup. Automatic and Zero Risk modes have independent connector identities, Tunnel IDs, API keys, runtime state, and recovery. Switching modes is transactional and never silently reuses the other mode's private credentials.
• Optional Zero Risk Pro profile. The install menu can add a second ChatGPT Web — Zero Risk Pro row with roughly 250,000 usable tokens before compaction. It is intentionally explicit: select ChatGPT Pro manually for every turn; a Pro account is required, and the larger limit may not work with every effort.
• Manual turns that remain real Codex turns. A neutral request ID binds each ChatGPT response to the exact Codex turn through MCP metadata. Concurrent turns stay isolated; reconnects rejoin the same request; cancellation, completion, retained-chat continuation, and compaction are explicit and fail closed.
• Zero Risk compaction. The active ChatGPT agent produces the checkpoint through the bound MCP control, the old chat is retired, and the compacted Codex continuation starts from a fresh manual chat without exposing private control values in the pasted prompt.
• A clearer launcher flow. First-run onboarding now offers With Automation and Zero Risk, defaults to automation, and keeps the same choice in Settings. The MCP guide includes trimmed, accelerated video walkthroughs with an in-launcher fullscreen viewer.
• Five-language launcher. Setup, mode selection, manual-turn guidance, diagnostics, and model-profile controls are available in English, Simplified Chinese, Traditional Chinese, Japanese, and Korean.

v5.0.7 hotfixes

• Compaction preserves the latest cumulative checkpoint. When an inline compaction request needs trimming, the newest checkpoint and final summary instruction are retained. Omitted history is disclosed; a checkpoint that cannot fit produces an explicit error instead of silent data loss. Fixes #447.
• Expired browser tasks release their runtime ownership. Removing an abandoned task now cancels its active request and frees the counters that could leave settings blocked. This addresses the cleanup defect in #426; the original large-context stall remains under investigation.
• Tunnel readiness no longer waits on an optional remote status lookup. Setup and launcher recovery use the tunnel client's local inventory and health endpoint. This fixes the reproduced network/proxy timeout from the macOS report in #438; it does not remove the network requirements for starting or reconnecting a tunnel.
• Response observation follows CSS visibility changes. Answers and completion controls revealed without a DOM mutation no longer remain stuck behind an empty cached snapshot. The defect was reproduced and fixed in Chromium while investigating #483; confirmation on the reporter's setup is still needed.
• Extra High and Pro availability are detected separately. An available Extra High setting is no longer rejected simply because Pro is unavailable. A disabled Pro row can also pass ChatGPT's own localized explanation to Codex. Addresses #416. Run Install into Codex again to refresh saved capability detection.
• Switching from Zero Risk back to automation publishes the correct browser target. Setup can inspect the selected mode's page, and a failed switch restores the previous descriptor. Fixes #482.
• Codex hook checks preserve unrelated TOML edits. Native config rewrites can place other tables between the managed hook and its trust entry; those entries are now checked separately without weakening integrity checks. Network/session verification errors also retain their actual diagnosis instead of reporting that you need to sign in. Addresses the reproduced cases investigated in #431.
• Native command approvals reach Codex intact. Full Harness forwards supported sandbox approval fields instead of dropping them. Codex still owns approval and execution. Fixes #487. Recreate the connector after updating if ChatGPT still has the old cached tool schema.
• Launcher startup failures stay visible and offer Retry. A bootstrap failure no longer leaves an unreachable background instance. Retry preserves saved settings and the ChatGPT profile. Addresses the invisible-launcher failure in #422.
• Localized response errors and diagnostics are clearer. Observed Stopped thinking labels are recognized across ChatGPT languages, including Chinese and Japanese, without treating quoted answer text as a failure. The launcher adds Traditional Chinese and Korean and fills gaps in native dialogs and diagnostics.
• Historical MCP handles are cleaned up after escaped newlines. Decoded context values are scrubbed while native tool-call IDs and JSON keys stay intact. New content-free MCP diagnostics distinguish calls reaching the server from protocol or result errors. Fixes #501; this is not a claim that upstream safety rejections are resolved.
• A cleaner README and support flow. The new monochrome overview keeps the existing demo, adds direct platform downloads, and makes setup, model access, and troubleshooting easier to find. Zero Risk guidance and the bug-report form are updated. Dependency security updates are included.

Thanks to @jamie950315 for the Traditional Chinese translation basis (#463), @H0ng for the Korean translation basis (#481), and everyone who supplied logs and reproduction details.

Please install the complete launcher update, reopen the launcher, and restart Codex before retrying the affected workflow. If the same failure remains, reopen its issue (or leave a comment asking me to reopen it) with your versions and a fresh Activity safe-log export from that attempt. Broader compaction, connector-discovery, and upstream safety reports remain open where their cause is not established.

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