GPT-5.5 and model clarity
The ChatGPT subscription provider now reflects the current GPT-5.5 experience instead of showing stale older entries. The header model picker keeps the full model name available, preserves important suffixes like Fast and Heavy, and gives long model IDs enough room to be read without turning the selector into a noisy panel. This matters most when switching between subscription models, OpenRouter models, and local models during the same session.
Remembered permissions you can manage
Tool permission prompts now show the actual request, target, command, and arguments in a clearer review surface, with sensitive and oversized values redacted or truncated. Allow once and always allow now behave differently by design: one-time choices apply only to the current prompt, while remembered choices are sent back to the backend as scoped permission rules. Settings now includes a Permissions tab where remembered allow and deny decisions can be reviewed, revoked one by one, or cleared in a single action.
Chat input and cost visibility
The composer now shows a compact cost hint before sending, and active sessions surface usage details more consistently while streaming. Input handling is safer for IME users: pressing Enter while composing Chinese, Japanese, or Korean text no longer submits the message. OpenYak tracks browser composition events, the legacy keyCode 229 path, and a short post-composition guard so finalized text lands in the input before Enter resumes its normal send behavior.
Streaming and error polish
This release also tightens the session event path. Usage stream events are isolated so a usage update cannot accidentally overwrite unrelated session state, active jobs become interactive earlier so permission prompts are less likely to race the SSE connection, and FastAPI validation arrays now render as readable UI errors instead of raw response objects. API error parsing is centralized so onboarding, provider settings, Ollama status, and context indicators speak with the same error language.
Full workflow validation
1.1.9 was tested through the desktop-style GUI flows behind GitHub issue #19: real ChatGPT subscription model selection, permission allow once versus always allow behavior, remembered permission review and revocation, cost/usage display, and IME composition submission safety. The release build passed backend permission tests, frontend lint, production Next.js build, Playwright preflight coverage, signed and notarized macOS Intel and Apple Silicon builds, Windows x64, Linux deb/rpm, and updater manifest publishing.
Upgrading from 1.1.8
Users on 1.1.8 will receive 1.1.9 through the in-app updater, or can download the installer from the Download page. No migration is required. Existing provider settings, conversations, workspaces, and local files carry forward; remembered permission rules start from the choices you make after upgrading.