OOMU® Control
AI agents your team can use. Policy and proof you can manage.
OOMU does the work on your team's Macs. OOMU Control is the server your organization runs to deploy it, decide which AI models every Mac may use, and see what each one is actually running and what it blocked. And if some of your people found OOMU before you did, that is fine: it is free at work, and Control brings those Macs under the same rules.
The honest post-mortem
It was never really about the model.
The rollouts that stalled did not stall because the model was not clever enough. They stalled for three reasons, and everyone involved knows what they were.
Nobody could see what it did. Legal asked what happened to the documents and nobody had an answer that survived a follow-up question.
Nobody would give it anything real. So it got the harmless work, produced harmless results, and justified nothing.
The bill grew faster than the benefit. Per-seat, per-question, per-region, plus the consultants brought in to explain the invoice.
There is a fourth reason nobody writes down. It was chosen at the top and handed down, so nobody below ever wanted it.
An agent nobody is allowed to audit is not an asset. It is a liability with a subscription.
OOMU Control
One place to answer the questions you are about to be asked.
Everything below is about the organization, not the person. Your people already have OOMU working the way they want it. Control is how you put a single shape around all of it without taking anything away from them.
- Your administrators sign in the way they sign in to everything else Your identity provider over OIDC or SAML, and the directory groups you already keep decide what each person can do. Disable someone and their sessions end at once.
- You decide which models are allowed Local models only, or local plus the cloud providers you approve. Add a corporate proxy and your own root certificate if your network needs them. Every policy version is signed with your organization's key, so a Mac can prove it came from you.
- Two people to loosen it, one to lock it down A policy author proposes a change and a different administrator approves it. The emergency air-gap is the exception: security can hold every Mac to local models at once, because that only ever makes things safer.
- Roles for the people who run it Super administrator, policy author, security operations, and auditor. The auditor can read everything and change nothing.
- See what each Mac is actually running Every Mac reports the policy version it applied, so you see who is current, who is behind, and who has stopped reporting, rather than assuming that publishing a policy means it arrived.
- Security events, where your people already look Blocked providers, policy updates, and enrollments, signed by each Mac and kept permanently. Streamed to Splunk, Datadog, Microsoft Sentinel, or a signed webhook, and exportable as CSV or syslog. Metadata only: the server refuses anything carrying prompt, response, or document text.
- Deploy it like software, not like a favor Control writes an expiring enrollment profile and install script for Jamf Pro, Kandji, Intune, Workspace ONE, or Apple Configurator, rather than you emailing everyone a link and hoping. Macs need macOS 14 and OOMU 0.3 or later.
- On your own infrastructure, with nothing phoning home One server with the console built in, on your own PostgreSQL, inside your network. The console loads nothing from outside it, and the license is a signed file with no license server behind it.
Not in Control yet: spending limits by team, and policy that reaches a Mac faster than its next check-in, which is every five minutes by default. They go on this list when they ship, not before.
Nothing that keeps a person safe is in that list, and nothing ever will be. Air-Gap Mode, the permissions macOS enforces, the encrypted history on each Mac, and its signed audit export all ship with the free app. Charging you for the safety would turn it into a sales tactic, and then it stops being safety.
Unit economics
The cheapest inference is the kind you already own.
Your organization has already bought a large fleet of capable computers. In most AI deployments they are used as thin clients pointed at somebody else's data center, and you are billed by the question for the privilege.
OOMU inverts that. Everyday work runs on the Apple Silicon machines you have already paid for, at no marginal cost. Cloud models stay available for the jobs that genuinely need them, on your keys and your policy, which turns spend into a decision instead of a surprise.
Budget stops scaling with curiosity. People are allowed to experiment again.
Licensing
The short version, said out loud.
Using OOMU is free, including commercially. An employee, a freelancer working for clients, a whole department that found it on their own. No license, no seat count, no letter from us.
OOMU Control is licensed. You pay when the organization wants to govern OOMU centrally rather than leave it to each person.
Selling OOMU on is a separate matter. Building it into your own product, reselling it, or running it as a service for other people are different rights from using it, and those are licensed too.
OOMU is proprietary software, licensed and not sold. Scope, terms, and support are agreed directly with Eldris.
Have your people try it first. That is usually the fastest evaluation there is.
It is free and it is the same product. Let them use it on something real, then come and talk to us about the fleet.