Not machine-to-machine sync.
Admin-to-fleet control.

mddb Fleet is for organizations rolling out AI agents across a team. The admin, or the admin's own AI agent connected over a dedicated admin-MCP, distributes mandatory skills and practices to every connected machine. Personal setups stay personal: the team layer applies on top, never by imaging machines into clones.

Talk to us about Fleet For individual developers

enterprise only · our cloud or your infrastructure

mddb fleet universe
admin agent, via admin-MCP
admin-agent → fleet.rollout { skill: "review-practices", version: "1.4.0" } fleet ✓ 17 machines scheduled (staged: 20% → 100%) admin-agent → fleet.compliance {} fleet ! 2 machines drifted: review-hook disabled locally (dev-linux-04, mac-07) # the admin's agent runs the fleet. no dashboard clicking required.

the problem

Rolling out agents is easy.
Governing them isn't.

// wild west

Every developer configures their agents their own way. Ten machines, ten different rulebooks, zero shared practices.

// practices that don't stick

The review checklist everyone agreed on lives in a wiki. The agents never read the wiki.

// shadow MCPs

Unvetted MCP servers and skills appear on work machines. Security finds out last.

// onboarding drag

A new developer spends days assembling the team's agent setup by hand, and still misses half of it.

// invisible spend

Tokens burn across the team with no per-machine picture: who spends what, and how much goes to actual work.

// no command center

There is nowhere to ask: is the mandatory skill active on every machine right now, or quietly switched off?

the solution

What the admin gets

Push, don't pray

Roll out a mandatory review-practices skill or an AI-manager reporting skill to the whole fleet, with versions and staged rollout.

Layers, not imaging

The team layer applies on top of each developer's personal setup. Nobody loses their environment; everybody gets the mandatory practices.

Compliance you can see

Is the client MCP alive on every machine? Are mandatory skills active, or quietly overridden by local settings? Drift is detected and visible.

Fleet inventory

Machine by machine: which entities, which versions, applied or failed. Onboarding and offboarding become one command each.

Admin-MCP

Your management agent connects over MCP and runs the fleet itself: distributes skills, checks inventory, reads compliance. The web panel is just another client of the same API.

Usage reporting

Work status and token spend per machine, including how much goes to actual work tasks. Transparent by design: devices always see what they report.

deployment

Your data, your choice of gravity

tier 1

On our infrastructure

A managed tenant on the mddb cloud. Fastest start, zero ops on your side.

  • Provisioned in a day
  • Backups, monitoring, upgrades included
  • SLA and priority support
tier 2

On your infrastructure

The same product, deployed on-premise or in your private cloud. Data never leaves your perimeter.

  • On-premise or private cloud installer
  • Commercial license, air-gap friendly
  • Long-term support channel

◆ integration is a separate line: SSO, SIEM, secret managers, custom harness adapters, migration and team training are scoped and priced as an engagement, not buried in the subscription.

principles

Control without creepiness

✓ devices always see what they report: telemetry is policy-enabled and visible ✓ no session content, no prompts, no code is ever collected ✓ enforcement by visibility, like MDM: drift is detected, people stay in charge ✓ the local agent is open source: auditable by your security team ✓ admin panel and admin-MCP use the same API: your agent is a first-class operator

One base. Every ship on course.

Pilot slots are limited while we build. Tell us about your team and be first in line.

Talk to us about Fleet