The In-App Assistant
The orb in the corner of every page — a chat that reads and edits your workspace
It acts — it doesn't just answer
The assistant is connected to the workspace you have open. Ask a question and it looks the answer up in your real segments, jobs, solutions, and research. Ask for a change and it makes it — creating, linking, and updating the same objects you'd edit by hand. Its replies link straight to what it touched: click Open in RoleDream on a step card and you land on that object in the app.
Because its edits are ordinary workspace changes, they show up for your teammates immediately and appear in the Activity feed like anyone else's work. Treat it as a fast colleague, not a sandbox: it can be wrong, so verify important changes on the canvas — the Blueprint is the review surface.
Documents are the best input
The fastest way to a mapped Blueprint is not typing — it's handing over what you already have: an interview transcript, a process document, an SOP. Use the + in the message box or drop files anywhere on the chat panel, then say what you want done with them.
They ride the next message
Attached files are sent with your next message only, then the queue clears — attach first, then write the instruction.
They stay with the conversation
Uploads are kept on the conversation (see its Files panel) and count toward your plan's storage. The workspace picker offers your earlier chat uploads, so a document is uploaded once and reused.
Say the tagging up front
"Map it as-is" or "draft it as to-be" in the same message — the assistant tags what it creates, and the Blueprint keeps current and future state side by side.
Conversations belong to a workspace
Every conversation is saved automatically and can be reopened from the history drawer (the clock icon in the chat header). The + next to it starts a fresh chat — the old one isn't lost, it's in history.
- swap_horizSwitching workspace starts a new chat. A conversation stays with the workspace it was held in — switch back and you'll find it in that workspace's history. Nothing leaks between companies.
- handymanChanging your tool set starts a new chat too. Enabling or disabling an Assistant Tool changes what the assistant can reach, so the conversation restarts rather than carrying stale tool context.
- hourglass_topBusy or rate-limited? The notice appears in the chat itself and your message stays in the thread — wait a moment and continue; nothing is lost.
What connectors let it do for you
Out of the box the assistant sees only this workspace. Each Assistant Tool you connect adds an outcome to the loop — the catalog groups them by exactly that, and the same grouping is the best way to think about what to ask for:
Understand the AS-IS
Bring the evidence in instead of retelling it. Interviews from Dovetail, docs from Notion or Confluence, real customer conversations from Intercom, CRM data from HubSpot, product analytics from PostHog — even the codebase from GitHub as the truest AS-IS of all.
"Pull my Dovetail interviews about onboarding and map what they say onto this segment's jobs."
"Check PostHog — do customers actually take these steps, in this order?"
Plan the betterments
A TO-BE that lives only on the canvas doesn't ship. With Linear, Jira, Notion, or Slack connected, the assistant turns the gap between AS-IS and TO-BE into tracked, owned work — and keeps the Blueprint and the backlog telling the same story.
"Compare the AS-IS and TO-BE for this job and create a Linear issue for every step that changes."
"Turn that Slack thread into a request linked to this solution."
Build the TO-BE
A well-described TO-BE flow is already close to a specification. Builders like Replit and Lovable generate running prototypes straight from it; Figma turns designs into them; Miro brings existing journey maps in; GitHub lets the assistant propose the actual code change.
"Generate a Replit prototype of the TO-BE flow — one screen per frontstage step, using the staged resources as the screens."
In-app assistant or your own AI tool?
Both work on the same workspace and both understand the model — use whichever is closer to where you're working, or both.
In-app assistant
Best while you're in RoleDream: mapping a document you just received, questioning a flow on the canvas, making a quick round of edits. Nothing to install, and every teammate with the feature has it.
Your own AI tool over MCP
Connect Claude, Cursor, or any MCP-capable assistant to RoleDream and the workspace joins the context you already work in — strongest for the Build step, where the TO-BE flow becomes code, prototypes, and tasks in your own environment. Set it up here.