DISPATCH · ANY SOURCE
your compute.
lumenflow governance.
Connect your own servers, GPUs and private infrastructure to the same control plane used for governed AI work.
One control plane, three sources
Wherever the compute runs, the policy, the evidence, and the record are the same.
Cloud models
Governed inference LumenFlow runs for you
BYOC — your hardware
Machines you control, governed the same way
LumenFlow compute
Managed execution, billed as compute credits
Governed inference
The simplest path is to let LumenFlow run the model call. We hold the provider keys, meter every call, and charge a flat governance fee on top of the token cost — drawn from your workspace’s inference credit. This is live today.
Bring your own key
Use your own model-provider accounts instead. Your provider bills you directly for tokens, and LumenFlow charges its governance fee for routing, policy checks, approvals, and the evidence trail wrapped around the call. Your keys stay in the cloud — they are never sent to a connected machine.
Bring your own compute
Run workloads on machines you control. LumenFlow’s cloud issues a scoped authorization envelope — permissions, budget, a kill switch, and a confinement posture — and a runner on your machine executes inside it, streaming evidence and cost back as it goes. Today that means connecting a Linux or Windows machine and keeping it running yourself as a terminal loop: connect, doctor, run. Linux gets LumenFlow’s strongest sandboxing by default; Windows nodes work today too, at a reduced isolation posture, and are kept off privileged work until that changes.
Your machine only ever connects outward, over HTTPS, to poll for work and report back — nothing inbound is ever opened to reach it. From your workspace’s Compute page you can drain a node before planned maintenance, quarantine one immediately if something looks wrong, or revoke its credential outright.
Keep data where it already lives
Governed work runs on your own hardware for residency or security reasons — nothing has to leave your network to be governed.
Use the machines you already have
Spare or specialised local compute does the work while LumenFlow's control plane still sets the rules and keeps the record.
The same evidence, every time
Every tool call streams evidence and cost back to the cloud, so a run on your own machine is as auditable as anything hosted.
Read bring your own compute for how the envelope and evidence chain work, or jump straight to setup: connect a Linux machine or connect a Windows machine. See also Connected Compute for how BYOC and LumenFlow-managed runners share one governance model.
Governed private compute
Beyond your own hardware, LumenFlow can run compute for you too — managed sandboxes, browsers, and execution runs on LumenFlow’s own infrastructure, billed as compute credits that are rolling out. It is a governed routing and execution substrate, with metering and checkout landing in slices, not a standing catalogue you shop from.
For workloads that need stronger data residency, evidence, or isolation than shared managed compute gives you, we build dedicated infrastructure for your workspace instead. Talk to us about your rollout.
See governed AI compute: campuses and data centres for what a dedicated site looks like at full scale.
From one node to a compute campus
The same governance model scales from a single connected machine to a dedicated facility.
Node
One machine you connect — a workstation, a server, a runner — that pulls governed work under your workspace's policy and reports evidence back.
Fleet
Several connected nodes in one workspace, each one drained, quarantined, or revoked on its own without touching the rest of the fleet.
Private cluster
Dedicated infrastructure held for one tenant, for workloads that need stronger isolation than shared managed compute gives you. Talk to us to scope it.
Compute campus
A dedicated AI data centre or GPU cluster — yours or a partner's — governed end to end: a sealed four-record proof bundle (utilisation, residency, carbon, SLA), residency enforced as a fail-closed gate on every event, and GPU attestation that's cryptographically verified, never silently upgraded from self-asserted. LumenFlow governs and proves what ran on that floor — it never operates the power, cooling, or physical scheduling behind it. Propose, never actuate.
Connect your compute
Pair a Linux or Windows machine today, or talk to us about dedicated infrastructure for your workspace.