Planner follow-through is stronger, personas stay consistent, memory stays current on long threads, and streaming + history replay are steadier across the extension and API.
What's new
Planner follow-through: Planner now does a better job tracking open loops (including reminders and due dates) so the agent can stay focused on finishing what you started.
Personas that stick: Grais can confirm your chat type once (business / friends & family / dating) and keep the right tone and next-step guidance throughout the thread.
Memory that stays current: Long conversations keep context aligned, with summaries that help the agent stay grounded without dragging the full backlog into every turn.
Steadier streaming + replay: Streams stay connected through idle moments, and refreshes preserve message/tool order so drafts, tool cards, and Quick Picks don’t jump around.
More channels supported: Grais now supports Tinder and LinkedIn, with chat switching that keeps drafts scoped to the right thread.
Why it matters
Open loops stay visible, so the agent can follow through on tasks, reminders, and next steps instead of dropping momentum.
Persona and memory improvements keep drafts aligned with the right voice without re-explaining context on long threads.
Steadier streaming and replay keep the transcript reliable across refreshes and different host sites.
Try it out
Start a new conversation, confirm your chat type when prompted, and ask for a draft—notice the tone stays consistent across follow-ups.
Ask for a plan with multiple steps, then add a reminder like “remind me tomorrow” and confirm it shows up as a due-dated task.
Start a longer reply, refresh mid-stream, and confirm the transcript rehydrates in the same order (including any tool cards).
Update the Grais Chrome extension, open Tinder or LinkedIn messages, and switch conversations to see the sidebar reset cleanly per thread.
If you already have access, you’ll receive this update over the next few days. If you don’t yet, join the waitlist to get access soon.
Behind the scenes
Prompting now prioritizes the most recent context and keeps Quick Picks short and actionable.
Planner and memory updates can run alongside replies, keeping responses fast while the agent stays grounded.
Streaming and stored history now share the same event framing and stable IDs, so replay matches what you saw in real time.