Over the last five weeks we shipped 1,100+ commits on speed and reliability. Three things matter most: a new desktop updater, a much better MCP, and a faster app.
A whole new updater
Some of you had Rize restart on its own, throw a blocking update window, or stop tracking after an update. Part of that was the updater. It ran on custom code we wrote years ago, and it had become fragile. We replaced it with a well-maintained updating library and rebuilt our release process around it. Sorry this took so long.
What changes
- Stable. Updates download in the background, verify themselves, and install on your next restart. Nothing crashes while you work.
- Lighter. The old updater kept checking and retrying in the background. The new one does not, so Rize uses noticeably less memory and CPU while it sits in your menu bar.
- A small notice, not a blocking pop-up. You restart when you want.
- Staged rollouts. A new version goes to 1% of devices, then 2%, 5%, 50%, then everyone. A bad build gets caught early.
- Faster fixes. We can ship a small fix in days instead of waiting for a big release.
What to do
- Restart Rize. You want 3.0.203 or newer. Check the Rize menu on macOS or the File menu on Windows and Linux. Behind? Grab the latest from app.rize.io/downloads and install over the top. Nothing is lost.
- Want features early? Turn on Receive Beta Updates, right under Check for Updates. Turn it off any time.
- Windows: the new installer puts Rize in a different folder, and Malwarebytes has flagged it for a few people. It is a false positive and we are working with them on it. Details here. If an update will not install, message us from the Help menu.
A much better MCP
Over half of you now use the MCP every day, from Claude, ChatGPT, Cursor or our own chat, so we spent the month reading real conversations and fixing what broke. Tool errors are down 13x: routines went from 12.4% to 0.5%, in-app tools from 9.6% to 0.7%. In the chats we reviewed, 95% now end resolved, up from 83% in October, and failures dropped from 8% to 3%. Whole error classes, like a date range the API would not take, are gone.
Batch tools. batch_create_time_entries fills a week from one message, with a dry run first. batch_update_time_entries retags everything that matches a filter. batch_delete_time_entries previews before it deletes. split_time_entry cuts one entry in two.
Ask why. get_time_entry can explain every tag: the rule, suggestion or past habit that set it, and how confident Rize was. Say "that project is wrong, fix it and make it stick" and Rize AI proposes the rule change and the backfill together.
Everything else from your agent. Create and run reports and routines. Draft an invoice from unbilled time and send it as a separate step. Ask "hours by member then project, this month versus last" with query_time_data. Export grouped CSVs. Set up a new workspace for an agency, law firm or accounting firm in one call.
More apps. Google Workspace, Linear, Atlassian, Attio, Slack, Fathom, Granola, Fireflies, Calendly, Loom, HighLevel, Featurebase and more, plus any custom MCP server. The 10-connector cap is gone. See Settings > Integrations.
Does what you asked. Give the details and Rize AI makes the change. One question max, only what you mentioned, a confirmation before anything is deleted, and it remembers what it already looked up.
Cheaper to run. Rize AI loads only the tools a step needs, caches what it reads, and sends less evidence to the model. Routine messages use 45% fewer tokens and weekly MCP token volume is down 30% from its peak, same usage. That is how we keep adding to the agent without touching your credits.
A faster app
We want Rize to feel instant, so we measured the pages you open most and went after the slowest parts. Your time entries now load in about 50 ms. The workspace sidebar, the day's meetings, tracking state and the app itself all load in well under 100 ms. AI label suggestions used to sit in a queue for up to 20 minutes at busy times; now they show up in under a minute.
The chart compares against last week's slowest point. Against a normal day, time entries and the sidebar are still about 3x faster. What we did, in plain terms:
- Fixed the slowest queries. One check ran five million times a day at 56 ms. Meeting lookups scanned half a million rows at 100 ms. Both are now about 1 ms, and total database time fell about 40%.
- More room on the database. Every query used to wait about 150 ms for a free connection. We quadrupled the pool, so they stop waiting.
- AI labeling runs 15 at a time instead of 3. The backlog drains about 5x faster.
- Timer start and stop on the web are instant. The screen no longer waits for the server.
- Background tabs stay quiet. A hidden tab stops refetching your day on every update and catches up when you come back.
- Tracking stays clean. Live "Tracking" entries close properly, and the widget no longer sticks on Syncing.
If anything in Rize still feels slow, tell us. Message us from the Help menu with what you were doing, and we will fix it.
There is more in this batch: the Integrations hub, the routine gallery, Meetings for everyone, the new review sheet and the Labels page each get their own post this week.
Thanks for sticking with us.
Macgill and the Rize team