What's new
- Live chat feels smoother: Streamed replies stay pinned more reliably, live text settles into markdown more cleanly, and loading states are less jarring while Grais gets the conversation ready.
- Startup and recovery are more reliable: Conversation lookup, replay startup, and auth return flows were hardened so Grais is less likely to get stuck after reloads, token refreshes, or older conversation state.
- Settings are clearer and more trustworthy: The settings flow is easier to scan, save and load problems surface more clearly, and model and assistant-behavior controls are easier to use.
- Transcript output is easier to read: Tool activity and file updates are summarized more clearly, with less raw config noise showing up in the conversation.
- Cross-channel conversation handling is tighter: Tinder startup now preserves the right identity and richer profile context, while conversation lookup behavior is more consistent across supported channels.
Why it matters
- A smoother stream makes Grais feel faster while a reply is still being generated.
- Better startup and recovery means fewer dead ends when you reopen the extension or return after login.
- Clearer settings and transcript output make it easier to trust what Grais is doing and adjust how it responds.
- More reliable conversation matching reduces context drift, especially in busy or long-running threads.
Try it out
- Open an active conversation and watch a streamed reply from start to finish. The transcript should stay steadier as the response grows.
- Reload the extension or return after an auth handoff and confirm the conversation restores cleanly instead of stalling on startup.
- Open settings and review the model and assistant behavior controls if they are available for your account.
- In Tinder or other connected channels, start or reopen a conversation and confirm the thread context is picked up correctly.
- No setup changes are required. These improvements are already live in the extension.
Behind the scenes
- Startup, replay, and streaming paths were tightened so loading and recovery behave more consistently.
- Settings, conversation creation, and transcript rendering now surface problems more clearly instead of silently masking them.
- Shared conversation handling was aligned across channels to reduce platform-specific drift.