The web app

app.matterwave.ai is the reference client: everything the platform can do is reachable from it, and every other client shows the same conversations.

Chat

• Streaming answers with a context meter and the model's name on every turn. A dropped connection resumes the stream where it left off.
• Branches. Editing a message or regenerating an answer writes a sibling; the original stays. Step between siblings at any turn. The terminal and editor clients open on the same active branch.
• Attachments. Text files and documents as parts of a message; what a model can read is shown with the model.
• Model switching per conversation, from the models your plan includes. A model outside your plan is shown locked with its upgrade path.
• Custom instructions that apply to every new conversation (how they work).
• Long conversations are compacted on the server so a thread can run past the model's window (compaction).
• Search, pin and archive across your conversations.

Settings

• Account: display name, the terms version you accepted, export everything, delete the account.
• API keys: create, scope (chat:read, chat:write) and revoke keys; the device-code sign-in the terminal and VS Code clients use issues keys that appear here too.
• Usage and budget: the allowance, what is spent, the overage cap, per-key usage.
• Organisation: members and roles, invitations, single sign-on on the Team plan, the plan and billing portal, the audit log.
• Devices: approve a device-code sign-in (how it works).

Security

Model output is treated as untrusted input: the markdown renderer never renders raw HTML, and the app ships a Content-Security-Policy behind it. The session is a cookie with a CSRF token; there is no analytics or advertising cookie.