Projects → Overview: rebuilt as a project manager's dashboard. Six readouts across the top (completed, in progress, to do, overdue, due this week, team members), a Project health card that puts the work done next to the time gone and says in one sentence whether the project is on track, and a Scope vs delivery chart that shows the backlog closing — or not. Below it: tasks by status and by priority, milestone progress, a Team workload list saying who is carrying what and who is late, and hours logged week by week. Tapping a readout opens the tab it is about.
Agents → Connect your AI: Claude now connects by signing in instead of pasting a secret link. You copy a plain address, add it in Claude, choose Sign in now, and Claude sends you to a NIZU page where you check what it may see and change and click Allow. Nothing secret ever leaves NIZU, there is no “is that you?” step, and Claude receives its tools the moment you allow it — no more Refresh tools.
Agents → Connect your AI: the “Connected” screen now tells you the two things still to do in Claude — switch NIZU on in a chat, and choose Always allow when Claude asks permission tool by tool.
Agents: an AI connected to NIZU can now manage the people of a client, not only the company. Four new tools — list, add, change and remove a client contact — follow the same rules as the Clients screen: a first name and a valid email are required, the email may not belong to a team member or to another contact of that client, the first contact becomes the primary contact, and a removed contact goes to the trash and can be restored. Adding, changing and removing are proposed for your approval like every other write. The Sales agent pack carries them by default.
Agents: seven new ready-made agent roles join the Project Manager WhatsApp shop assistant, Sales, CEO brief, CFO, Developer, Marketing assistant and an Administrator that can use every capability its role allows. Picking a role now pre-fills the capabilities it needs.
Updates
Projects → Overview: the project description has taken the place the activity log used to occupy — the widest card near the top — and the activity log moved to the bottom of the tab, where a history belongs. The facts everybody asks for (status, client, dates, price, labels, the estimate or order it came from, and your own custom fields) are now one list in a single Key facts card.
Projects → Overview: money at a glance for the projects you can see invoices and expenses for — project value, invoiced, paid with what is still outstanding, and expenses.
Projects: a project's tabs and the tab you are on are now one card — the strip sits on top of the panel it opens instead of floating above a second box, and the Overview finally sits in the same panel as every other tab.
Projects: the Settings button above a project is gone. Its three options (can the client see the timesheet, Slack notifications, task statuses to leave out) are now a Project settings section at the bottom of Edit project, so everything about a project is edited in one place and the button bar is no longer crowded.
Projects: the buttons above a project (Reminders, Actions and everything in it) now use the Font Awesome 7 icon set like the rest of the app.
Agents: the Claude instructions follow the current Claude app step by step (Profile → Settings → Connectors → + → Add custom connector → Sign in now), including what to do on a Team or Enterprise plan and if you chose “Sign in when connected”.
Agents: connecting an AI by signing in now needs the same permission as the Connect your AI page. A team member without it, or a client contact, is refused on the sign-in page instead of being able to connect from the AI's side.
Agents: the NIZU sign-in page that Claude opens recognises a connection you started from the wizard, shows the choices you already made, offers “Everything I can do” like the wizard does, and returns you to Claude after Allow — or, on Cancel, tells Claude you said no instead of leaving you on the dashboard.
Fix
Agents (MCP): an AI client connected through an Agents connection could not create a client, project, task, comment or team member — the call failed with Attempt to read property "id" on null, because nobody is logged in during an MCP call. The record is now created on behalf of the account the connection acts as (the connection owner, or the agent's own user), and the AI gets the new id back.
Agents: clicking Allow on the sign-in page that Claude opens did nothing — the page simply came back. The connection could not be completed this way and a second connection row appeared. Fixed.
