Codex Web GPT v5.0.2

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

v5.0.2 hotfixes

• Zero Risk continues after automatic compaction. The new browser turn gets its own context-epoch identity instead of reusing an already-completed turn. An ordinary final answer that finished just before compaction remains available, and reconnecting does not submit the task twice. Fixes #318.
• The older post-compaction turn_id conflict is addressed. Codex can continue an existing instruction under a new native turn ID without adding another human message. The bridge now recognizes that continuation only when it matches the exact completed checkpoint, thread, turn, model, effort, and source instruction; cancellation and mismatched identities still fail closed.
• Resumed tasks recover their actual native environment. Root tasks, not only subagents, can recover their current working directory and permissions from their exact Codex-owned rollout when the wire context is absent or historical. A fresh context-only environment after compaction must agree with that rollout before tools can run. Automatic and standalone compaction use their explicit current/source turn identities, never an arbitrary older workspace. Extends the recovery covered by #288.
• Zero Risk compaction allows two minutes for the manual handoff. The compacting prompt and its Sent-to-connector-start step now each allow 120 seconds. Ordinary Zero Risk turns keep their 30-second deadlines.
• Bigger Context compaction has balanced parts and separate progress budgets. Ordered messages stay intact, the largest staged part is minimized, and each part is checked against the selected account's per-message limits. Each acknowledgement has a bounded wait; accepted staging progress renews the handoff budget, and the model-generation window starts after the final compact prompt is accepted. The requested work effort is restored after staging. Addresses #316.
• Repeated connector verification cleans up correctly. Verification preserves an already-selected connector while checking it, then clears the editor through native selection/deletion instead of leaving a stale mention or plugin draft. Visible personalization controls remain detectable during hydration. Verify still does not submit a task or create a turn token: start a real Codex task with a Web model afterward. Addresses #293 and the verification flow in #315.
• Browser diagnostics keep structural evidence without rendered UI text. Default browser-state traces no longer store page titles, conversation URLs, sidebar/menu text, or connector-row text. Screenshots require explicit full-diagnostics opt-in.
• Regression checks and cleanup. Added native continuation/permission checks and Zero Risk compaction coverage, replaced a timing-sensitive test with controlled clocks, and removed redundant selector paths and source-format assertions.

Included fixes from v5.0.1

• Interrupt now retires the exact complete turn. Stopping a Codex turn cancels its matching HTTP stream, browser execution, MCP capability, and structured compaction owner. An interrupt that arrives before ownership registration is also remembered, and a replacement waits for physical teardown instead of inheriting a ghost turn. This fixes the stuck-next-message lifecycle reported in #307.
• Zero Risk outcomes remain truthful. A launcher or runtime failure is no longer relabeled as a user cancellation merely because its own cleanup retires the turn capability.
• Zero Risk → Automatic switches no longer require a launcher restart. The already-loaded browser surface receives its ownership marker inside the same setup transaction, while a failed switch restores the exact previous runtime and mode. Fixes #306.
• Model-switched subagents recover only trusted Codex environment. If Codex omits the complete environment block, the bridge can recover the exact current child rollout from Codex-owned state and canonical JSONL. It validates lineage, turn identity, workspace paths, and sandbox permissions and never treats user-authored XML as filesystem authority. Addresses the remaining case in #288 and the same v5.0.0 spawn failure reported in #314.
• Unrelated Codex configuration is no longer mistaken for bridge drift. Reconnect checks compare only the exact keys managed by Codex Web GPT, preserving unrelated [agents] and [features] settings. Fixes #311.
• Bigger Context uses its real 3× ceiling once. Multipart staging no longer multiplies an already-expanded context limit again, and a Pro rate-limit dialog that appears immediately after submission is now returned as a structured limit error.
• Retired MCP work cannot leak into the next turn. Capability retirement rejects stale tool boundaries, clears obsolete completion vetoes, and settles the previous browser owner before a replacement begins.
• Failed connector verification now leaves actionable evidence. Verify runtime writes an ordered local checkpoint trace of the Temporary Chat surface, composer state, menus, and connector rows while keeping composer text out of the trace. The launcher also records the exact verification failure in Activity, so intermittent reports such as #293 can be diagnosed instead of guessed at.
• Connector setup troubleshooting is explicit. The guide now covers account-matched Tunnel IDs and API keys, tunnel readiness, and the one-time connector retry that ChatGPT may require after its first handshake.

Addresses #288, #306, #307, #311, and #314.

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