New in early access — Remote Devices

AI agents that can work
wherever your data lives.

Raigent connects governed AI agents to tools, workflows, and remote computers across labs, company sites, clouds, and on-prem environments — all controlled from one place.

No inbound ports. No arbitrary remote access. Devices connect out — you stay in control.

Built for environments where compute doesn't live in one cloud

On-prem data centers
Lab & instrument machines
Edge & field sites
Multi-cloud
Restricted networks
// The gap

Frameworks build agents. Clouds host them. Nobody connects the real world.

There are dozens of frameworks for building local agents and plenty of managed clouds for hosting them. But your compute, data, and instruments are distributed across sites, networks, and trust boundaries — and that's exactly where agents struggle to reach.

Compute is fragmented

A GPU box in a lab, a workstation behind a factory firewall, jobs in two clouds. Each is isolated, each needs its own access story, and none of them talk.

Networks fight you

On-prem and lab machines sit behind NAT and firewalls with no inbound access. Today that means VPNs, bastion hosts, and tickets — brittle plumbing that breaks at the worst time.

No single pane of glass

Once agents are running in five places, you lose the thread: who called what, which tools are reachable, what's failing. Security and ops fly blind.

// How it works

One control plane. Agents, tools, and devices across every environment.

Remote workers and device services poll out to Raigent. The control plane applies tenant-scoped policy, dispatches durable work, and keeps operations visible — no inbound firewall rules required.

Raigent Control Plane Policy · Durable execution · Audit Console · SDK · MCP agent lab-gpu-01 agent cloud-east tool site-db device lab-pc-17 agent poll ↗ ↖ poll
01 · ENROLL

Connect the environments you already have

Register agents, tools, and remote computers to a tenant. Each device receives its own identity, queue, and deliberately narrow policy.

02 · POLL

They call out, not the reverse

Workers poll Raigent over outbound connections. No open inbound ports, no VPN, and no general-purpose remote shell exposed on the network.

03 · BROKER

The control plane authorizes every job

Raigent resolves the tenant, target, workflow, and allowed operation before work reaches a remote environment, then tracks the durable execution.

// Remote Devices

Give agents controlled access to remote computers.

Enroll a lab PC, workstation, or on-prem server once. Raigent agents can then discover the device, inspect approved paths, list directories, and move an approved file to managed object storage.

01

Admin-controlled enrollment

Create, rotate, revoke, and monitor tenant-scoped device credentials from the Raigent console.

02

Operator-approved access

Paths are resolved and checked against configured roots. The first release supports metadata, bounded directory listings, and file uploads — not arbitrary commands or file mutation.

03

Files take the direct route

Uploads travel from the selected device to an allowlisted HTTPS destination using a short-lived signed URL; file bytes do not pass through the agent or workflow history.

Remote Devices
tenant / research
PC
Lab PC 17
device/research/lab-pc-17
ACTIVE
“Find the latest results in /lab/runs and make the CSV available for analysis.”
list-remote-devicesfound
list-remote-directory12 entries
upload-remote-fileverified
Tenant and device identity are platform-derived. The agent supplies only the selected device ID and an approved path.
// The platform

Everything you need to run agents across trust boundaries

One control plane for policy, connectivity, and observability — so distributed agents behave like a single, governed system instead of a pile of scripts.

Outbound-only connectivity

Agents and tools reach the control plane with outbound polling, so they run behind NAT and firewalls with zero inbound exposure. Reach machines you could never reach before — without opening them up.

Policy-driven mesh

Declare exactly which agents and tools may communicate, in which direction, for which operations. The control plane is the single point of enforcement — change a rule, the mesh reconfigures.

Durable workflows

Run interactive, scheduled, and multi-step agent work on durable execution. Requests keep their identity and state as they cross service and network boundaries.

Cross-cloud & on-prem

The mesh spans clouds, regions, and your own data centers as one fabric. An agent in AWS can safely use a tool that lives on a machine in your lab — brokered, logged, and governed.

Console, SDK & MCP

Manage agents, skills, tools, workflows, schedules, mesh policy, and devices in one console — then integrate with applications through the SDK or MCP.

Framework-agnostic tools

Bring agents built with any framework and expose any capability as a tool. Raigent doesn't replace how you build — it's the connective tissue that lets those agents work across sites.

// In practice

From enrollment to useful work in a few controlled steps

  • Create the device in the Raigent console and issue its one-time credential.
  • Enroll the service without putting secrets in command history.
  • Keep the worker connected in the background while the desktop window is closed.
raigent-device — secure enrollment
# paste the one-time key into a hidden prompt
$ raigent-device enroll --config /etc/raigent/config.json
Device credential: ••••••••••••
✓ credential stored securely

# run as a background service
$ raigent-device start
✓ service started

$ raigent-device status --config /etc/raigent/config.json
device/research/lab-pc-17 · connected
0
inbound ports to open
1
control plane, full visibility
1:1
scoped identity per device
3
scoped device file operations

Bring your agents to the work — without opening your network.

Raigent is in early access. Tell us where your agents, tools, and remote computers need to work, and we'll help you connect them safely.