Product / Needs You & Governance
Your AI team knows when to ask.
You decide what AI can do. MasterMNG keeps work moving and brings you the decisions that need a person, shows you why you are being asked, and holds back what you did not approve wherever MasterMNG runs the work.
One queue for everything that needs a human.
Instead of watching every agent, you watch one list. Needs You gathers what only a person can resolve, across the whole company, and links each item to its source.
- Approvals, decisions and reviews.
- Blocked work and failed runs.
- Handoffs addressed to a person.
- Budget limits, missing credentials, permission requests and runners that went offline.
- Operating rules proposed by agents, waiting for confirmation.
Decisions and approvals that hold at the server.
Decision Desk
An agent that needs you files a request with its reason, its options, the budget and time impact and the evidence, then stops. You approve, reject or ask for a revision. The agent is woken exactly once with your answer, and the decision is checked again when its effects are applied.
Approvals bound to the exact operation
An approval covers one specific operation — its task, run, targets, parameters and environment — for a limited time. It is used once. A changed, replayed, expired or cross-company use is refused and recorded.
Delivery gates
A task can carry a delivery contract. It cannot be marked done until a delivery is accepted, bound to the exact version of the work delivered, with an explicit human review confirmation.
Rules confirmed by people
Standing operating rules change how agents behave, so they take effect only after a person confirms them. Every change is versioned and recorded in the activity log.
Explainable autonomy: what an agent may do, and why.
Live in production. Governance policies and approvals appear in the Company Brain, and the Live Map shows what needs judgment, what is expiring and what is blocked.
Effective capabilities
One computation of what an agent may do on a given task and run — from company and department policy, its tool profile and its skills — recorded so you can see the result and where it came from.
Hard tool permissions
Where MasterMNG executes the call — the tool gateway, project tools and the MasterMNG runtime — a denied tool call never executes, whatever a prompt, skill or document says. The server is the boundary, not the instructions.
Deny wins
A block from company or department policy, or from an agent's tool profile, cannot be overridden by a skill, a narrower rule or a task. Conflicts it cannot resolve fail closed and go to a person.
Server-side approvals
A withheld action runs only after a person approves that exact action, and policy is checked again immediately before it runs. One approval, at most one execution.
Safe expiration
A request nobody answers expires and is never executed. Expirations are recorded and shown, not hidden.
Three lanes in Needs You
Needs your judgment; authorized by policy, without asking you; and expired safely. You see what was decided for you, and by which policy.
Policy provenance
Every allow or deny names the policy that produced it, so an automated outcome can always be traced back to a rule a person set.
Honest enforcement state
Each control says where it is enforced, partially enforced or only advisory — for example on a customer's own machine, where the provider's tool applies the limits. Advisory is never shown as enforced.
Human attention load
Measures of how much judgment the company asks of people, how much policy handles, and how much expires — so you can tune autonomy with evidence.
Get started
Build your AI workforce.
Create your account and set up your company, or tell us what you want to run and a person replies.