What the Rize Agent Does
The Rize agent answers questions about your own tracked time in plain language. It reads your time entries, projects, clients, tasks, labels, and teams, then replies with the breakdown and the entries behind it.
It runs inside Rize, so there is nothing to connect. It uses the same toolset the Rize MCP server exposes to Claude and ChatGPT, except it already knows who you are and which workspace you are in.
Opening the Agent
The agent lives at the top of the sidebar under Home > Agent. From any screen, Cmd+J on macOS or Ctrl+J on Windows opens a new chat.
The panel opens beside whatever you were looking at. You can keep a report or your timesheet on screen while you ask about it.
Starting From a Skill
A new chat offers four presets, so you do not have to write a prompt to get something useful:
- Review My Day -- a summary of everything you worked on today
- Analyze Productivity -- your patterns over the past week
- Generate Weekly Report -- a weekly rundown of time and activity
- Review Recent Work -- a pass over recent activity and time entries
Each preset sends a short question and attaches a fuller recipe behind the scenes, so the thread stays readable while the answer stays specific.
You are not limited to the presets. Anything you would ask a colleague works: "How much did I bill to Acme last month?", "What did Tuesday afternoon go to?", "Which project ate the most meeting time this week?"
Saving Your Own Skills
A skill is a prompt you save once and reuse. The Skills button in the chat composer lists the ones available to you, and picking one drops it into your message instead of making you retype the recipe.
Skills have a visibility: personal, team, or workspace. The composer shows team skills first, then workspace, then personal. Create and manage them in Settings > Agent.
Saved skills are readable from the Rize MCP server too, so a recipe written once works in Claude or ChatGPT as well as in Rize.
Answers That Link Back to Your Data
When the agent names a project, client, task, label, team, member, or time entry, it renders as a pill instead of plain text. Hover for details, click to open the record.
This is what makes an answer checkable. Ask what your time entries look like today and the breakdown shows the project attached to each block, so a stretch of work sitting on the wrong client is obvious rather than buried in a number.
Anything you do not have permission to see stays as plain text. Pills resolve only against records your account can already open.
Chat History
Every conversation is saved and reopenable from the history list in the panel. Chats are titled automatically from your first message, and reopening one restores the full thread including any files you attached.
Conversations created by a scheduled report or routine are kept out of the list, so history stays what you actually asked.
Attachments and Dictation
You can attach up to 4 files per message: images, PDFs, and text files, capped at 25 MB in total for one message. Drag a file onto the composer, paste a screenshot, or use the attach button.
Dictation is on the microphone button. The transcript appends to whatever is already in the composer, so you can start typing and finish out loud.
Minimizing docks the chat at the edge of the screen and keeps the thread alive. Switch pages, check a number, reopen, and carry on.
Chatting About a Report
A chat can start attached to something rather than from a blank page. Opening Chat on a finished report run carries that run's period, scope, and filters into the conversation, and a chip at the top of the thread shows what it was started from.
That is the fastest route from "what happened" to "why":
- "Why were meeting hours up this week?"
- "Which entries drove the Acme number?"
- "Compare this to the same week last month."
See AI Reports in Rize for how to build the reports these chats hang off.
Tips
- Name the period. "This week", "last month", and "yesterday" resolve against your own timezone.
- Name the entity. Referring to a client or project by name beats describing it.
- Ask follow-ups. The thread keeps its context, so "now break that down by label" works without restating the question.
- Use it to fix data, not just read it. The agent can create and update entries, approve or reject pending suggestions, and retag work.
- Reach for MCP when the question spans tools. The Rize MCP server puts the same data in Claude or ChatGPT alongside your other sources.