Your network,
operated by agents.

The agent-native control plane for DDI and DCIM. One live twin of your estate, on open standards, with a deterministic Rust math core underneath. Signed lease ingestion and tenant isolation work today; the agents that would act on the twin are in development.

omnitwin design preview the agent workflow, not a live session
$ omnitwin agent launch scout --subnet 10.0.0.0/16
✓ sweeping 65,534 addresses  [████████████████] 100%
✓ 1,284 hosts found · 47 undocumented

$ omnitwin drift detect --scope production
✓ comparing intent to reality  [████████████████] 100%
! 12 records diverged from intent

$ omnitwin reconcile --plan drift-8f2a --apply
✓ applying 12 changes  [████████████████] 100%
✓ twin aligned · 0 conflicts · audit logged

$ _

Speaks the protocols you already run

DNSDHCPIPAMDCIMBINDKeaNetBoxTerraform

Five surfaces. One twin underneath.

Discovery, allocation, drift, pre-flight and audit all read and write the same live model of your estate, so an answer from one is an answer the others already agree with. Each surface below says plainly whether it ships today.

Find the hosts nobody wrote down.

A sweep reads the estate directly rather than trusting the record of it, then reports the three numbers that matter: what answered, what the IPAM already knew about, and the gap between them.

In development
  • Addresses swept 65,534
  • No record anywhere 47
discover design preview
$ omnitwin discover --subnet 10.42.0.0/16 --probe icmp,snmp,arp
✓ sweeping 65,534 addresses  [████████████████] 100%

  responding            1,284
  already in IPAM       1,237
  undocumented             47

$ omnitwin discover adopt --undocumented --dry-run
✓ 47 hosts would be written to the twin
✓ 0 collisions with existing records

$ _
10.42.0.0/16 1,284 responding

Four layers, milliseconds to cross them.

A request never lands in one place. Plan reads it and dispatches, the intent store and the reality store answer at the same time, and the math core waits for both before it returns an address.

plan

step 1

> PrefixService.AllocateNextFreeSubnet

pool10.42.0.0/1665,536 addresses

size/2662 usable hosts

policyfirst free

dispatchintent, realityin parallel

waiting on both

intent · postgres

step 2 · parallel

> reservations held in 10.42.0.0/16

10.42.0.0/24256 addresses

10.42.1.0/24256 addresses

10.42.2.0/25128 addresses

10.42.3.0/2664 addresses

4 blocks held704 addresses

reality · graph

step 2 · parallel

> observed state for 10.42.0.0/16

blocks cross-checked4

in-flight leasesnone observed

conflicting edgesnone observed

drift against intentnone observed

clear to allocate

math core · rust

step 3

> first_free(pool, /26, held)

0110.42.0.0/26heldby 10.42.0.0/24

..8 moreheldby 3 reservations

1010.42.2.64/26heldby 10.42.2.0/25

1110.42.2.128/26free

→ 10.42.2.128/26

A design preview of the trace, not a recorded session. The gRPC method and the arithmetic are real and every address is computed; the agents that would drive this path are in development.

A different architectural bet.

Agent-native, not agent-wrapped

The design puts agents beside the data rather than calling out to it: persistent state, identity-bound access, no context re-hydration per call. Agent interfaces are in development and no agent is wired yet.

Built for the new buyer

Designed for the engineer with a credit card. A $25 entry tier that is actually useful, and a self-serve path all the way up.

A twin that acts

Drift is the gap between what you intended and what the network is doing. Surfacing it, and reconciling it on one approval, are in development. The what-if check that tests a change before production ships today.

Pricing that implodes the premium tier

$1,250/month against legacy platforms billing six figures a year. Same capability, different foundations.

Find what your records never knew about.

Passive telemetry only sees a device when it decides to talk. Scout is designed to sweep the subnet directly over ICMP, ARP and SNMP, so silent bare-metal, static assignments and unallocated hardware all surface. Device discovery is not built yet, and the session below is a design preview.

  • Would run on a schedule, or on demand after provisioning.
  • Infers device type without an inbound firewall hole.
  • Writes straight into the twin, with no manual entry.
See the agent guides
scout design preview
$ omnitwin agent launch scout --subnet 10.20.0.0/16
  probing icmp · arp · snmp

✓ 10.20.4.17    dell poweredge r760   undocumented
✓ 10.20.4.18    cisco nexus 9300      matches twin
✓ 10.20.9.204   silent bare-metal     undocumented

  1,284 hosts · 47 undocumented · 12 stale
safety-net design preview
$ omnitwin apply plan.tf --simulate
  cloning twin branch sim/8f2a

✓ vlan 340 delegation resolves
✓ dhcp scope headroom 38%
✗ 10.128.0.0/20 overlaps sn_01HX7

  blocked before production. 1 conflict.

Break it in simulation, not at 3am.

The what-if check tests a proposed change against your twin, in memory, before production sees it. Overlapping CIDRs, exhausted scopes and broken zone delegation get caught by hard mathematical rules, not by a reviewer. This one ships today, as ordinary code rather than as an agent. The command below is a design preview: the tenant CLI is still in development.

  • Fails the push instead of failing the network.
  • Runs in CI next to the rest of your pipeline.
  • Every simulation is auditable after the fact.
How the twin works
Drift detected

When operational state diverges from the source of truth, the canvas is designed to show you, and to reconcile on one approval with a record of who changed what. Drift detection is in development.

Why OmniTwin.

Built different from the foundations up, not bolted onto a record store.

Works with your stack

OmniTwin sits alongside the DNS, DHCP and CMDB systems you already run. No rip and replace: your infrastructure stays yours and gains a live twin of itself.

Real-time, not reactive

Agents catch drift the instant it happens and stage the reconcile. No stale spreadsheets, no 3am page for a mismatch that was preventable.

Zero-risk validation

Changes run through a sandboxed twin before they reach production. Mathematical proof that a change cannot cause an overlap, a leak or a policy violation.

An order of magnitude cheaper

Legacy DDI platforms bill six figures a year on rigid licences. OmniTwin starts at $25 a month with the same enterprise capability.

What we believe.

Open-standards DDI is not just architecture. It is a global repository of networking practice, stress-tested by thousands of engineers who face the same complexity you do.

Open standards

Unlike proprietary tools built by generalist vendors, OmniTwin was forged in the trenches of real network operations, with agents that act on their own.

Real operations

DDI should not be a spreadsheet your team dreads updating. It should be a living twin that discovers, detects drift and reconciles.

Autonomous DDI

Notes from the build.

View all posts

Claim your code.

The beta is now open. Every signup gets an access code: it holds your seat in a cohort and puts you first in line for launch pricing and unlocked features.