AI project management that knows what your projects cost

Most AI project management tools can summarise a task list. Ask them whether the engagement starting Monday will make money and they have nothing to answer with, because nobody told them what an hour of that team costs. SUNAGO Matrix connects the assistant you already use to the system that holds the real numbers.

Start a free trialSee how it connects

Two things are called AI project management

The first is a language model attached to a task board. It writes the status update, summarises the comment thread, drafts the plan. Every major project tool ships some version of it now, it genuinely saves time, and it never touches money. Ask it what the project is earning and it will either decline or guess.

The second is an assistant with access to the system of record: the budgets, the logged hours, the cost rates, the invoicing plan. That one can be asked the question a partner actually has, which is not "what are the tasks" but "are we making money on this, and if not, when did it start going wrong".

Matrix is built for the second. There is no chatbot inside the product, and that is on purpose: you already have an assistant you trust, and it is better than anything we would bolt on. What we provide is the connection, so it can read and update your workspace as you, with your permissions.

Things people actually ask it

Each of these is answered from live data, not from a summary written last week.

  • Which projects are running over budget right now, and by how much?

    Reads budget, spent to date and forecast at completion for each project.

  • Who is allocated above 100 percent over the next three weeks?

    Reads the allocation plan across every person and project.

  • Log six hours on the Novo Nordisk integration for Tuesday, non-billable.

    Writes a real time entry, costed at that person's real rate.

  • What is in my pipeline that has not moved in a month?

    Reads leads and deals, active ones by default rather than every closed record.

  • Which fixed-fee milestones are due but still not marked invoiced?

    Reads the billing plan. One of four ready-made prompts on the server.

  • Add a risk to the Aarhus rollout: the client's test environment is not ready.

    Writes to the project risk log.

The assistant works out which of the 105 available tools it needs and in what order. Four ready-made prompts ship with it for the routine questions: what is waiting on me, project status, log time, and ready to invoice.

An assistant is only as honest as the rate underneath it

Most systems derive an employee's hourly cost by dividing annual salary by a nominal year, something like 1,850 or 2,000 hours. Nobody bills 1,850 hours. Take out the vacation entitlement, account for part-time capacity, then multiply by the utilisation the person actually achieves, and the number of hours you can charge against that salary is far smaller. Add their share of company overhead, including colleagues who bill nothing, and the true cost of a billable hour moves a long way from the naive figure.

This is why the first kind of AI project management is so comfortable. A model reading a flattering margin will report a flattering margin, fluently, and nobody in the room will question it because it came from the system.

Matrix derives the rate the harder way as standard, and every margin, Estimate at Completion and Forecast at Completion in the product is built on it. A consultant running at 60 percent utilisation costs materially more per billable hour than the same salary at 85 percent, and the projects they are on say so while there is still time to act. The arithmetic is written up in full in true cost per hour for consulting firms.

Connecting it takes about a minute

The connection uses the Model Context Protocol, an open standard for letting an assistant work with an external system. It is not ours, which is why the list of clients that support it keeps growing. There is nothing to install, no API key to create and nothing to configure on our side.

Claude, web and desktop

Add a custom connector and paste the address below. Claude sends you through the normal Matrix sign-in.

https://sunago-matrix.com/mcp

Claude Code

One command, then /mcp in a session to sign in.

claude mcp add --transport http sunago-matrix https://sunago-matrix.com/mcp

Cursor and VS Code take the same address in their own config file. The full instructions for every client, the complete list of 105 tools and the permission model in detail are in the public documentation repository.

Once it is connected, ask it to call whoami. It answers with who you are signed in as and what you are allowed to see, which confirms the connection works before you rely on it.

What it can see, and what it cannot

Connecting an assistant to the system that holds your project economy is a decision about trust, so here is the shape of it in specifics rather than reassurance.

It runs as you

Every call carries your own sign-in and reaches the database as you, under row-level security. There is no shared key and no administrative bypass, so it cannot reach a workspace you do not belong to.

Your permissions still apply

Each tool checks the same permission the Matrix interface checks. Without it the tool refuses and names what is missing, so you know which admin to ask rather than seeing an empty answer.

Most of it only reads

31 of the 105 tools cannot change anything at all. Every write that does happen is recorded with the tool, the caller and the arguments.

Money is gated separately

4 tools read financial data behind the finance permission and 4 can write, and those writes are billing milestones and a project's invoicing settings. Nothing on the server issues an invoice, sends one, or records a payment. No tool reads or writes an annual salary, a gross hourly rate, a national ID, or any bank or card detail.

Deleting is deliberately hard

Deleting a lead, contact, deal, project or task is restricted to workspace admins and reversible from the recycle bin for 30 days. 10 of the more consequential tools refuse to run until a human supplies an exact confirmation phrase, which the agent is told in as many words that it must not invent.

Two capabilities go further than the rest and we would rather you read it here than discover it: the cost per hour derived from salary is returned per named employee to anyone who already holds the reporting or time approval permission, and an administrator with HR edit rights can grant or remove workspace admin rights through an assistant. Both are the same authority that person already has in the interface. The full model, tool by tool, is in the permissions documentation.

AI project management, answered

The questions people ask before connecting an assistant to their project data.

What is AI project management?
In practice it means two quite different things, and the difference matters. Most of what is sold as AI project management is a language model bolted onto a task list: it drafts status updates, summarises comments and suggests a schedule. That is useful, and it is not project management, because none of it touches what the work costs or earns. The second kind connects an assistant to the system that holds your real numbers, so it can answer questions about money rather than about tasks. That is what SUNAGO Matrix does. It has no chatbot of its own; it opens your workspace to the assistant you already use, under your own permissions.
Will AI replace project managers?
No, and the reason is worth being precise about. What an assistant is good at is the part of the job that is really data retrieval: which projects are drifting, who is overallocated next month, what has not been invoiced. Those questions have correct answers and a person currently spends hours assembling them from spreadsheets. What it cannot do is decide what to do about the answer, tell a client the scope has changed, or judge whether a team can absorb another engagement. The realistic effect is that a project manager stops being a reporting function and goes back to managing.
What are the best AI project management tools?
The honest answer is that the category is very young and most of the tools in it are doing the first kind of AI project management described above: summarising and drafting inside a task tool. If that is what you need, the major project tools all ship some version of it. The question worth asking of any of them is a different one: can it tell you whether a project is making money, and what is the cost rate behind that answer. If a tool derives an hourly cost by dividing salary by a nominal year, its margins are flattering and its AI will repeat them confidently.
Can AI see our real project costs, or just hours?
It sees the same figures your finance view shows, and they are built on a real cost per hour rather than salary divided by 1,850. Matrix derives the rate from capacity, vacation entitlement and the utilisation a person actually achieves, then carries it through every margin, Estimate at Completion and Forecast at Completion in the product. That is why an assistant connected to it can answer whether a project is worth continuing, which is not a question a task tool can be asked.
Is it safe to connect an AI assistant to our project data?
It runs as you. The connection uses your own sign-in, and every call reaches the database as your user under row-level security, checked against the same feature permissions the Matrix interface checks. If you cannot see project finances in Matrix, an assistant connected as you cannot read them either. There is no shared key and no administrative bypass. Of the 105 tools, 31 only read. Deleting one of the five top-level records is restricted to workspace admins and reversible for 30 days, and 10 of the more consequential tools refuse to run until a human supplies an exact confirmation phrase. Every write is recorded with the tool, the caller and the arguments.
Which AI assistants does it work with?
Any client that supports remote MCP servers, which is the open standard for connecting an assistant to an external system. Claude on the web and desktop, Claude Code, Cursor and VS Code all do today, and the list is growing because the standard is not ours. There is nothing to install and no API key to create: you point the client at one address and sign in.
What does it cost?
There is nothing to buy to switch it on and calls are not metered. What an assistant can actually reach follows your workspace's own feature permissions, so the answer to what you can do through it is the same as the answer to what you can do in Matrix. If you are unsure what your workspace includes, ask us.

Ask your assistant something you cannot answer today

The trial includes the whole product. Set up your projects and your people, connect the assistant, and ask it which engagement is quietly losing money.

Start a free trialBook a demo