Managed Matrix Cloud
Matrix operates a hosted computer and platform services. Review workspace regions, model-provider routes and connected services.
Compare deployment choicesYour infrastructure · Self-host preview
Bring agents, files, terminals, internal apps and a database into one persistent Matrix computer. Evaluate the manual self-hosting preview on a compatible Linux server, with browser and CLI access.
Self-hosting is a preview. Full local inference and air-gapped operation require additional integration and qualification.

A useful AI task often needs more than a conversation. It needs source files, tools to work with them and somewhere to keep the output. Matrix brings those parts into a persistent computer your team can return to.
The self-host preview is for builders prepared to operate that environment. You own access protection, DNS, TLS, patching, backups, upgrades and incident response. Review the installation prerequisites and responsibilities before starting.
Matrix operates a hosted computer and platform services. Review workspace regions, model-provider routes and connected services.
Compare deployment choicesOperate the browser workspace and CLI on compatible Linux infrastructure. Managed Desktop/mobile pairing and platform integrations are not included by default.
Check the preview scopeBring the exact data boundary, identity needs and workflow. The preview does not establish complete local inference, enterprise governance or disconnected operation.
Review the required profileRunning Matrix on your server does not automatically keep prompts and files away from external AI providers. The selected agent, model, document processor and tools each need an approved route.
For a deployment without internet access, the complete workflow must work inside that boundary: installation, identity, inference, updates and recovery. Ask what has been demonstrated on the intended build rather than relying on the word “private.”
Read Matrix’s current data-flow and deployment boundariesIllustrative evaluation: supplier-document review. Use approved specifications, test reports and a requirements list to prepare a missing-document queue and draft review pack. A named engineer checks the references before anyone sends a message or changes a system.
This is a proposed evaluation, not a customer result or a preconfigured industry product. Agree the permitted data flows, expected artifact, quality standard and recovery checks before connecting live data.
Yes. A manual preview installer is documented for a compatible Linux server. Browser and CLI are the documented standalone paths. You operate and protect the server.
No. Processing depends on the selected feature, agent, provider and configuration. External providers may receive task context. Local inference needs separate qualification across the required workflow.
The current self-host preview does not establish full air-gapped support. A supported profile needs complete installation, internal identity, useful task execution and recovery evidence.
The runtime is available under AGPL-3.0-or-later. Review the current source and license, plus the terms for separate components and services.
You provide infrastructure and operating work. Model serving, support and third-party services can add costs. Hosted plan prices are a different offer.
Review the source and license or download the cost worksheet.
Tell us what your team wants AI to do, which data cannot go to the available services and who will review the result. We’ll identify what can be evaluated now and what needs more work.