Global Assistant
The Chat tab at the top of the sidebar is a global AI assistant. Unlike the chat inside a project, which only sees that one project's files, the global assistant works across everything in your workspace and can reach outside it: the live web, academic papers, and more.
Open it from Chat in the sidebar, or go to /chat. It greets you with a blank composer. Just ask.
What it can do
Ask in plain language. Depending on what you ask, the assistant can:
- Search across all your projects. Find a quote, a file, or a finding without remembering which project it lived in.
- Search the live web. Pull in current information and outside context, then read specific pages to quote them.
- Search academic papers. Look through peer-reviewed work via Semantic Scholar and OpenAlex when you need grounded sources.
- Summarize your account. Totals across projects, interviews, surveys, and files.
- Build charts. Bar and pie charts render inline in the reply.
- Build artifacts. Working HTML pages or React components that render live, with Preview and Code tabs.
- Create or open a project. Spin up a new workspace, or jump into an existing one.
Each conversation is saved as a thread. History reloads when you reopen a thread, and a live timeline shows what the assistant is doing while it works.
Connectors
The + button next to the composer opens Connectors, where you choose which sources the assistant may use:
- Web (on by default): search and read the live web.
- Repository (on by default): your projects, files, and transcripts.
- Academic (opt in): scholarly papers.
- Reddit (paid plans): posts and discussions.
Turn a source off and the assistant won't reach for it. That's useful when you want an answer grounded only in your own files, or only from the web.
File uploads
The same + button lets you attach files to a message: up to 5 files at once, 500 MB each. Uploads are transcribed or text-extracted first, so you'll see them process before you can send. They're saved to a Chat uploads folder in your media library, so you can reuse them later.
Deep Research
On paid plans, switch on Deep Research to turn one question into a full investigation. The assistant breaks the question down, searches every source you've enabled (your files, the web, academic papers), checks what it finds against itself, and returns a long, cited report.
It's built for depth, not speed, so a run can take a minute. You can run up to 10 a day.
Plans and limits
- Free: 20 assistant messages, total. After that you'll be asked to upgrade.
- Paid plans: unlimited messages, plus Deep Research and the Reddit connector.
Messages are counted per account across all your threads, so the Free allowance covers everything you ask here. This is separate from the per-project chat, which has its own limit.
Tips
- Say which sources you want. "Search only my files" or "check the web" steers the connectors without opening the menu.
- Be specific about scope. "Across the onboarding study and the pricing study, what did people say about trust?" beats a vague prompt.
- Use Deep Research for the big questions and normal chat for quick lookups.
- Follow-ups stay in context. Push back if an answer is too general.