Teams operating in classified and disconnected environments have long faced a hard tradeoff: adopt modern AI tooling and give up control of your data, or keep control and fall behind.
We built our latest release to end that tradeoff. Here's how it works, and what it means for teams running Mattermost in air-gapped networks.
Local models, zero egress
Everything runs inside your boundary. Models are deployed alongside your Mattermost instance, and no token ever leaves the network:
- Self-hosted inference — bring your own model server or use our reference deployment
- Full audit trail — every prompt and completion is logged to your compliance store
- Role-based access — administrators decide which teams and channels can invoke AI
What early adopters are seeing
"We cut incident summary time from forty minutes to four, without a single byte leaving our enclave."
That pattern repeats across defense, finance, and critical infrastructure: the value of AI shows up fastest in high-tempo operational workflows — incident response, shift handoffs, after-action reports.
Getting started
The AI plugin ships with the current release. Install it from the marketplace, point it at your model server, and grant access per team.
For deployment guidance in classified environments, see our deployment documentation or talk to your account team.