v6.1.29

Fame Codex 6.1.29

• Include the selected TH AI Pass model ID when creating an isolated temporary chat. Previously, only the subsequent send-message request carried that model selection, which could let the new chat start with the upstream default.
• Add regression coverage for temporary-chat creation and switching between Gemini, DeepSeek, and Claude models across consecutive isolated turns.

The bridge now carries the selected model through both conversation creation and message submission. The upstream service's actual model selection remains unverified: live requests returned HTTP 502 during diagnosis. A model's self-reported identity alone is not reliable verification.

After updating Fame Codex, reload the TH AI Pass browser extension, keep a signed-in TH AI Pass tab open, then start a new Codex session.

v6.1.28

Fame Codex 6.1.28

• Fix TH AI Pass reasoning-effort selection: expose each available chat model's advertised thinking levels in Codex, and forward the selected effort to the browser bridge. Upstream max maps to Codex Extra High.
• Preserve per-model thinking metadata across extension discovery, launcher state, and the Codex model catalog. Models without advertised thinking controls continue using the website default.
• Model refresh now reports loader failures and empty results instead of treating a fallback or stale model list as fresh. The extension refresh button reports HTTP failures.
• Show advertised thinking levels in TH AI Pass settings and document how account eligibility and text-only filtering affect the visible model list.

After upgrading, reload the TH AI Pass Chrome/Edge extension, open a signed-in TH AI Pass chat tab, refresh models in Fame Codex Settings, then restart Codex and open a new session. Model and effort availability may differ by TH AI Pass account.

v6.1.27

Fame Codex 6.1.27

• Intel Macs are no longer supported. macOS releases are Apple silicon (arm64) only. Existing Intel installs keep working but will not receive updates, and the installers refuse Intel Macs.
• Add Reset embedded browser data in Settings. It clears cookies, cache, and site data in the embedded ChatGPT browser; you sign in again and your regular browser and launcher settings are unchanged.
• The embedded browser now follows the launcher language after Fame Codex is fully quit and reopened. ChatGPT may still use its own saved language setting.
• Make reading and selecting the ChatGPT reasoning-effort menu more reliable.
• Recognise sub-agent lists in Codex environment context when recovering a chat's working folder.

Versions 6.1.25 and 6.1.26 were never published; this release supersedes them.

After updating, fully quit Fame Codex from the system tray and reopen it so the runtime restarts. Trusted Codex homes added in 6.1.24 take effect at that restart.

v6.1.24

Fame Codex 6.1.24

• Add Settings → Trusted Codex homes. Choose other Codex homes, such as an Orca account, so ChatGPT chats can recover their working folder (cwd) from sessions stored there. Each folder must contain a sessions folder, at most seven can be added, and the change applies the next time the launcher-owned runtime starts.
• TH AI Pass: retry a malformed tool call up to twice before reporting an error, enforce single-call turns when parallel tool calls are disabled, and give clearer messages when the browser extension is disconnected, timed out, or its tab is sleeping.
• TH AI Pass: the extension prefers a tab that is not frozen and reports a sleeping tab instead of waiting for it.
• TH AI Pass prompts now tell the model that the supplied tools act on the real workspace and not to claim work a tool result does not confirm.

After updating, reload the TH AI Pass browser extension. To use an Orca account's history, add its Codex home under Trusted Codex homes, then restart Fame Codex.

v6.1.23

Fame Codex 6.1.23

• Read the completed TH AI Pass loader record without waiting for the HTTP stream to close. This restores model discovery and temporary chat creation when the site keeps its stream open.
• Prefer the active or most recently used TH AI Pass chat tab, instead of an older retained tab that may be unresponsive.

Live tests on the installed bridge completed a text reply and a read-only file tool call with both Gemini 3.5 Flash Lite and Claude Opus 5.5. TH AI Pass remains experimental, and sleeping browser tabs can interrupt requests. Keep a signed-in TH AI Pass tab open and exclude the site from browser sleeping tabs.

After updating, reload the TH AI Pass browser extension and open a new Codex session. On Windows, exit Orca and Fame Codex before installing if Orca is holding the old application files open.

v6.1.22

Fame Codex 6.1.22

• Accept the text.verbosity preference sent by Codex. TH AI Pass now expresses low, medium, or high verbosity as a final-answer instruction instead of rejecting the request before contacting the model.
• Replace HTML-like tool-call delimiters with nonce-bound line markers. This avoids the extra closing angle bracket observed in live Gemini 3.5 Flash Lite tool responses. Tool names, arguments, nonce, and complete frames remain strictly validated before any call is exposed to Codex.
• Preserve the original cancellation reason when a macOS passkey-login quit command is aborted, including cancellation while the command is running.

Live validation with Codex and TH AI Pass Gemini 3.5 Flash Lite completed both an exact TEST_OK reply and a real read-only terminal command (Get-Content package.json -TotalCount 1), followed by the model's READ_OK response using the command result. This verifies one model and these flows; it does not guarantee every listed model or tool.

TH AI Pass remains experimental. Upstream policy refusals, the local 256K prompt budget, text-only input, and unsupported remote compaction still apply.

After installing, open a new Codex session. If the Windows installer reports that it cannot uninstall the old application, close applications holding Fame Codex's installation files, including Orca when it has loaded resources/app.asar. Finish active work before quitting either application.

v6.1.21

Fame Codex 6.1.21

• Add Settings → ChatGPT models → Detect models again. Detect available models from the signed-in account, save the result, and refresh the running bridge without restarting it. Failed detection preserves the previous settings. Open a new Codex session to load the refreshed model catalog.
• Increase the experimental TH AI Pass local prompt budget from 16K to 256K tokens. Count the complete outbound prompt with a tokenizer instead of estimating from raw request characters. Preserve instructions, tool schemas, and conversation history; existing 16K settings migrate automatically.
• Report TH AI Pass policy refusals explicitly without exposing the private prompt echoed in upstream errors.

Known limitation: TH AI Pass can still reject a Codex context with HTTP 422, CHAT_RAI_VIOLATION (Content Policy Violation). This release fixes the local budget rejection; it does not resolve upstream policy refusals or guarantee coding-agent compatibility. The 256K budget is a local limit, not a claim about any upstream model's capacity. Remote compaction remains unsupported.

Install through Update when you are ready to interrupt active work. Updating restarts Fame Codex and temporarily disconnects its bridge. After the app reopens, start a new Codex session. Do not remove your plugin or delete account data to refresh models.

v6.1.20

Fame Codex 6.1.20

Integrates upstream 6.1.18 and its test-stability follow-up.

• Recognize current ChatGPT model controls. An unrecognized interface now fails explicitly instead of overwriting a paid account's model catalog with Luna-only capability.
• Preserve current workspace roots, goal edits, task identity and completed work across partial context updates and compaction.
• Recognize previously observed ChatGPT messages when they return into view, and retain more precise stream-completion and network-error diagnostics.
• Import macOS passkey sessions only after the dedicated Chrome process exits cleanly.
• Keep session reminders above the embedded browser and improve update-error reporting and retry handling.
• Preserve Fame Codex's compact interface, Sleep/Wake, TH AI Pass, connector names and separate DEV profile.
• Adopt the upstream Temporary Chat setting. Existing Save chats preferences migrate to the equivalent setting; an explicitly configured Temporary Chat preference takes priority.

After installing, run Setup → Install into Codex → Reinstall, then fully quit and reopen Codex to refresh Web models. Do not delete launcher data or recreate the plugin just to refresh the model catalog.

The upstream model-selection and environment reports still under investigation are not claimed as resolved by this synchronization. Account limits and model availability still apply.

v6.1.19

Fame Codex 6.1.19

• Add Sleep / Wake in Settings. Sleep restores the previous Codex routing configuration without removing the launcher, tunnel credentials, or ChatGPT session. It remains asleep across launcher restarts. Wake reconnects the bridge. Restart Codex after either action.
• Fix model detection for the current ChatGPT model picker. A missing retired button attribute no longer causes Sol-capable accounts to be recorded as Luna-only. Run Setup → Reinstall, then restart Codex to refresh the model list.
• Recheck saved tunnel credentials when opening MCP setup and before reconnecting. Show the credential fields when saved credentials are missing, and report a specific recovery message.
• Keep the patched minimatch version for current build tools while preserving the older API required by legacy packaging dependencies.

Sleep restores only settings managed by Fame Codex; it does not delete the entire Codex configuration. The ChatGPT plugin and tunnel remain installed.

v6.1.18

Adds experimental TH AI Pass support alongside ChatGPT Web. Windows and Linux builds are included. TH AI Pass requires an enabled browser extension and a signed-in session. Codex CLI end-to-end tool use remains under testing. macOS builds are not included in this release.