Local-first agent infrastructure

AI agents that never
leave your building.

Agents Incorporated runs autonomous AI agents entirely on hardware you own. No cloud round-trips. No data egress. No third party reading your inbox to answer a question about your own filesystem.

operator@vm500-test — agentctl

    
The platform

Three pieces. All of them yours.

Every component runs inside your perimeter. Pull the network cable and the agents keep working.

Agent Runtime

A sandboxed execution environment for long-running agents. Deterministic tool calls, hard resource ceilings, and a full audit trail of every action taken.

Fleet Control

Schedule, version and roll back agents across every node you run. Declarative config, gradual rollout, and instant kill switch on any agent in the fleet.

Model Vault

Signed, content-addressed model storage. Pin exact weights per agent, verify provenance on load, and roll the whole fleet forward when you choose to.

How it works

Running in an afternoon.

No forklift migration and no rearchitecting. It installs beside what you already run.

Install the runtime

One binary or one Helm chart onto existing hardware. Bare metal, VMs, or any conformant Kubernetes cluster.

Mount your context

Point agents at the filesystems, databases and internal APIs they should see. Everything else stays invisible to them.

Define the guardrails

Declare allowed tools, rate ceilings and escalation rules. Denied by default — an agent can only touch what you named.

Ship it

Roll agents out per team or per node, watch the audit log, and revoke any one of them in a single command.

Why local

The tradeoff nobody talks about.

Hosted agents are convenient right up until you read what leaves the building.

 Hosted agentsAgents Incorporated
Where your data goesSomeone else's datacentreNowhere
Median tool-call latency380 ms12 ms
Works with the WAN downNoYes
Per-seat pricingForeverNever
Audit trail you controlPartial, on requestComplete, on disk
Model changed under youWhenever they shipWhen you say so
Running live

This page is the demo.

Everything below is read from the cluster serving you right now — not a screenshot, not a mock.

Reading cluster…
Served by pod
Node
Pod uptime
Cloudflare edge
Protocol
TLS
Origin round trip
Your IP
Two replicas sit behind this hostname. Hit Refresh a few times — the pod name changes as the load balancer moves you between them.
By the numbers

Built for the boring parts.

0
Bytes egressed
0ms
Median tool call
0+
Bundled tools
0%
On your metal

Minimum node

4 vCPU · 8 GiB RAM · 20 GiB disk
Linux 5.15+ · containerd or Docker
GPU optional — CPU inference supported

Integrates with

Kubernetes · Nomad · systemd
Postgres · S3-compatible object stores
OIDC / LDAP · Prometheus · OpenTelemetry

Keep your data where it is.

Start with one agent on one node. Grow to a fleet when it earns the trust.