Skip to content

v5.0.0

Latest
Latest

Choose a tag to compare

View all tags
@miuuyy miuuyy released this 03 Sep 14:11
· 2 commits to main since this release
v5.0.0
b2793cf

Codex Web GPT v5.0.0

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.

Important

If you use this project, please read Discussion #188. A short X video or mention showing a real Codex workflow helps keep this method from being reduced to “abuse” and lowers the risk of broad restrictions.

Zero Risk Mode running through the native Codex harness

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.
  • Three-language product parity. The new setup, mode selection, manual-turn guidance, diagnostics, and model-profile controls are complete in English, Chinese, and Japanese.

v5.0.0 hotfixes and reliability fixes

  • Accepted responses are no longer lost during a DOM stall. After ChatGPT accepts a prompt, the launcher can rebind the exact tab and continue observing the same response without resending it; MCP activity remains authoritative while the page recovers. Fixes #285.
  • Bigger Context final sends keep their full bounded stage budget. The 180-second stage deadline now remains the sole send budget instead of being silently collapsed by Playwright's inner 20-second press timeout. User cancellation and fail-closed ambiguous-acceptance handling are unchanged. Fixes #302.
  • Per-turn browser bootstrap is ordered correctly. A new tab is not leased to the browser helper until its exact internal idle document and ownership marker are committed, removing the ERR_FAILED (-2) race whose later symptom was an invalid or expired turn token. Fixes #295.
  • Current effort popovers are activated structurally. The launcher handles the current slider/menu shape, re-resolves replaced composer nodes, and keeps selection fail-closed instead of producing an empty item list after a forced click. Fixes #268.
  • Pro + Plan mode keeps its MCP turn alive. Tunnel and runtime supervision now distinguish real active work from idle state, preserve the exact managed owner, and recover the verified loopback MCP endpoint instead of entering a false reconnect loop. Fixes #283.
  • Codex Voice bypasses the Responses-only bridge. Setup journals and restores the official WebRTC call route separately from openai_base_url, so Voice session creation no longer falls into a local endpoint that cannot serve it. Fixes #258.
  • Resumed model-switched tasks recover a trusted cwd. When Codex omits <cwd> but supplies trusted workspace roots, the bridge uses the first validated root and keeps the existing absolute-path and containment checks. Fixes #288.
  • Windows shortcut activation is deterministic. A second normal shortcut launch now restores and focuses the existing launcher window across the single-instance boundary instead of leaving only a background process after reboot. Fixes #287.
  • Windows broker transport keeps its proven ownership boundary. Ordinary named-pipe calls resolve only after physical close, while explicitly unbounded tool waits settle on their complete framed response. This avoids both Bun's abortive-teardown crash and a stuck Zero Risk long-poll.
  • Safer cancellation and compaction ownership. Shutdown and cancel-all include structured compaction owners, wait for physical browser/helper settlement, evict failed retained handoffs, and never report an idle runtime while owned work is still active.
  • Manual-send recovery is explicit. The launcher preserves the prepared prompt and exact turn outcome across navigation replacement, keeps cancellation available after Sent, and returns a terminal error instead of creating or replaying an ambiguous chat.
  • Mode-aware setup and repair. Existing v4 Automatic credentials migrate only to Automatic; Zero Risk stays separate. Reinstall, disconnect, uninstall, and failed mode switches preserve or restore the exact previous Codex route, connector, credentials, and runtime ownership.
  • More useful Pro-limit diagnostics. Only the exact item index 4 (min=0; max=3) failure now explains that recent heavy Pro use may have temporarily hidden Pro after its usage limit was reached. This is diagnostic guidance, not a fallback or a claimed picker fix; #300 was closed after the limit behavior was identified and handled in v5.0.0.
  • Release and support polish. The README has a shorter Full Harness path, Zero Risk documentation, video-linked troubleshooting, current architecture and release checks, generated local artifacts are excluded from source commits, and packaged runtime integrity remains transactionally verified.

Addresses #258, #268, #283, #285, #287, #288, #295, and #302.

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
Assets 21
Loading