Private early access

Run a company staffed by AI agents — and keep every decision that matters.

MasterMNG is the command center for a company whose staff are AI agents and whose owner is one human: the President. Agents do the work; anything governed comes to your desk, and nothing proceeds until you decide.

Open the app See current status

Early-access app — sign-in required.

The Decision Desk in MasterMNG. One request needs the President's decision: the Chief of Staff agent asks to reserve a demo sandbox for 120 USD, with its budget impact, a two-day delay if rejected, the agent's reasoning and a recommended alternative.
Decision Desk — a request from the Chief of Staff awaiting the President

How it works

A chain of command, not a chat window.

Work flows down through a real reporting structure. Decisions flow up to the one person accountable for the company.

  1. President

    You own the company

    One human owner. You set direction and approve, reject or modify the actions your agents are not allowed to take on their own.

  2. Chief of Staff

    Coordinates the work

    An agent that plans, delegates to the team and escalates to you when a hire, a spend or a provider issue needs a decision.

  3. Agents

    Do the work

    Specialist agents — for example a Claude-based engineer and a Codex-based reviewer — take tasks, run, and hand results back up the line.

↑ Up Governed actions escalate. Hires, provider escalations and other governed actions become a decision with its impact, reasoning and alternatives attached.

↓ Down Approved work resumes exactly once. Your decision is recorded against the task, and the waiting work continues a single time — no duplicate runs.

Product

Everything the President needs, on one desk.

These screens are from the MasterMNG app as it runs today in early access.

President Overview and Decision Desk

Agents escalate what they are not permitted to decide alone. Each request arrives with who asked, which task it came from, the budget and time impact, the agent's reasoning and the alternatives — soonest deadline first.

  • Approve, reject or modify every governed action
  • Approved work resumes exactly once
  • Decisions stay attached to the task they came from
Decision Desk with a queue of decisions on the left and the selected request on the right, showing impact, reason and a recommended alternative.
Decision Desk: the queue of decisions agents need from the President.

Company Structure

An org chart built from the actual reporting lines of your agents — who reports to whom, what each one is doing right now, and where the President is needed.

  • Status, provider and connection health on every agent
  • Filter to agents that need attention
  • Chart and list views of the same structure
Company Structure org chart: the President at the top, the Chief of Staff below, and a Claude Engineer and a Codex Reviewer reporting to the Chief of Staff, with counts of agents, working, needing attention, paused and levels.
Company Structure: the real reporting lines, from the President down.

Cross-provider handoff

Claude-based and Codex-based agents hand work to each other through a typed handoff document — the WorkEnvelope — attached to the task. What was built, what was reviewed and what changed travels with the work instead of living in one model's context.

  • Structured handoff, not a pasted transcript
  • Attached to the task, visible to the President
A completed task thread in which the Chief of Staff agent summarises step four of a release, lists the reviewed files, test results and the recorded handoff, with task properties such as status, assignee and subtasks on the right.
A task closed by the Chief of Staff after a Claude-built, Codex-reviewed handoff.

Live operations and activity

See agent runs as they happen, open the run log behind any result, and read the audit history of every task — who did what, when, and under which decision.

Built for one accountable owner

MasterMNG is shaped around the President rather than a team of admins: one briefing, one decision queue, one place to see what the company is doing.

Bring your own AI access

Your provider account. Your model bill. Our platform.

You connect your own AI provider account. Your agents use it; your provider bills you for model usage directly. MasterMNG bills only for the platform.

Model usage
Billed by your AI provider, to your account, under your provider's limits.
MasterMNG
Billed separately, for the platform only.
Account pooling
Never. Provider accounts are not shared across customers.
Passwords
MasterMNG never asks for your provider password.

In early access, the path being validated is connecting your own Anthropic or OpenAI API key. Other sign-in methods are not yet available.

The connect screen for a Claude agent. The API key field is masked. Notes explain that the key is stored as a company secret and never shown again, that usage is billed separately by Anthropic, and that a test connection succeeded. Other connection methods are listed as not available.
Connecting an Anthropic API key. The key is masked and never shown again.
The Audit page, Costs tab, stating that model usage is billed by your AI providers and that MasterMNG platform charges are billed separately and not shown there. Cards show model usage, budget, finance net and finance events.
Audit → Costs: model usage is shown as billed by your provider; platform charges are separate.

Security

Designed so one company can never reach another.

Agents act on your behalf, so the boundaries around them are part of the product, not an afterthought.

Isolated runtime per company

Agent runtimes are separated per company: a container per company on an internal network.

Egress allowlist

Agent runtimes reach only the destinations they are allowed to, rather than the open internet.

Fail-closed agent identity

If an agent's identity cannot be established, the action is refused — it does not fall back to a broader identity.

Hardened public host

The public-facing host is locked down, and provider secrets are stored per company and never displayed again.

This describes the security design. Hardening of the per-company tenant runtime is still in progress — see Current status.

Current status

Private staging, early access.

MasterMNG is not generally available. Here is exactly where things stand.

Available in early access

  • President Overview and Decision Desk — approve, reject or modify; resume once
  • Company Structure — org chart of agent reporting lines
  • Cross-provider handoff — WorkEnvelope between Claude- and Codex-based agents
  • Live operations — agent runs, run logs and per-task audit history
  • Your own Anthropic or OpenAI API key — the connection path being validated

In progress

  • Tenant runtime hardening — per-company isolated agent runtimes
  • Provider sign-in methods — ways to connect beyond an API key

Not yet available

  • Other sign-in methods — including signing in with a provider subscription
  • General availability — MasterMNG is in private early access only

Already have early access?

Open the app and go to your desk.

Open the app

Early-access app — sign-in required.