Network automation your whole team can operate.
Not just the one person who writes the scripts.
A network admin with zero programming knowledge wires visual nodes into a complete operation — config backup, OS upgrade, compliance audit, VLAN and ACL provisioning — then runs it on demand, on a schedule, or straight from a monitoring alert. Dry-run diffs, approval gates and auto-rollback are on by default, because it is a production network.
Bring one multi-site workflow and one non-negotiable trust requirement — we map both in the session.
Flow Builder
device-upgrade
Validate
Run
Approval Gate
The flow is paused. Nothing on the device changes while it waits.
against real devices
The problem
One person scripts. Everyone else waits.
The runbook lives in someone’s head
That, and a folder of Python nobody else will touch. When they are on leave, the change waits — or someone brave does it by hand at 2am.
Every vendor is a rewrite
Arista, Junos, SR Linux and Cisco each get their own script, their own edge cases, and their own quiet way of breaking eight months later.
Nothing is provable afterwards
Who approved it? What actually changed on the box? A screenshot pasted into a ticket is not an audit trail, and the auditor knows it.
WeftRoute turns all three into one canvas your whole team can read, run, and answer for — with the credentials and the execution still inside your own environment.
The product
Low-code, end to end.
Every capability is a palette node: collect, parse, decide, change, verify, notify. Parsing uses TextFSM, regex or YANG — no AI required. Nothing falls back to “write a script for that part”.
Every arrow can branch on the previous node’s result, so a flow handles the failure path as explicitly as the happy one. Run it on demand, on a schedule, or from an external trigger — a monitoring alert, an ITSM ticket, a CI pipeline.
Fleet health, run ledger with per-node output, compliance posture, and a “waiting on you” approvals queue.
Probe, back up, diff and restore any device directly — the same safety rails as a flow, one device at a time.
Eleven node categories, one canvas
One flow, many domains
Open a Jira change ticket → gate on a change window → canary-upgrade one branch switch → verify the whole site still converges → close the ticket.
One graph. One audit trail. Your service desk, your switches and your servers are all first-class node categories — not integrations bolted on the side.
Safety rails
It is a production network. The rails are on by default.
You do not opt in to being careful. Every destructive capability ships behind a preview, a gate, a window and a way back.
Dry-run diff preview
See the exact config delta before anything is applied — and the approval gate shows that same diff to the person signing off.
Human approval gates
A named human clicks in the run ledger. The flow waits as long as it takes, and not one byte changes on a device while it does.
Change windows
Block or defer writes outside maintenance hours. On a routine flow like guest-WiFi PSK rotation, the window is the control — no approver needed.
Auto-rollback
A failed apply reverts to the pre-change state on its own. VM snapshots, database migrations and container updates each carry their own way back.
Reload-gated upgrades
A device is never rebooted unless you explicitly opt in. Free-space gated, image verified, downgrade-guarded, hardware-validated per platform.
Git-versioned history
Every config version committed to Git, with one-click restore — previewed first, like every other change on the platform.
AI is an accelerator, never a dependency
Describe a flow or a parser in plain English and WeftRoute generates it — validated, then saved as ordinary editable data you can read and change. Execution is deterministic with no model in the loop. The whole feature is off by default.
Templates
Twenty flows that already work.
Every one is a real, safety-railed graph you open and edit — not a blank canvas and a tutorial. Each already carries its own backup, preview, gate and rollback.
Vendor support
Multi-vendor by architecture, not by promise.
Transport — SSH, Telnet, NETCONF, RESTCONF, gNMI, SNMPv3, vendor HTTP APIs — is kept separate from capability behind one stable gRPC contract. A new platform is an addition, never a redesign.
| Platform | Facts | Backup / restore | CLI | Apply | OS upgrade | YANG |
|---|---|---|---|---|---|---|
| Arista EOS | ✓ | running + startup | ✓ | dry-run + rollback | staged | all three |
| Juniper Junos | ✓ | running + candidate | ✓ | candidate / commit | staged via PyEZ | NETCONF |
| Nokia SR Linux | via gNMI | canonical JSON | ✓ | — | — | gNMI |
| Cisco IOS / IOS-XR / NX-OS | ✓ | ✓ | ✓ | ✓ | staged | transport-ready |
| Your platform | A plugin scaffold maps your existing Python device code onto the contract — zero core edits. | |||||
Green marker = verified live against real hardware. Staged upgrade = image transferred and boot set; the activation reboot only happens when you pass reload explicitly.
Security & the trust boundary
We run the control plane. You keep the credentials and the runtime.
Hosted coordination should not require hosted custody. The boundary is built into where each responsibility runs — not into a policy document.
Credentials never enter the SaaS
Profiles and secrets resolve only inside your Edge, at the moment local execution needs them — from your Vault, as static secrets or one-time SSH credentials and certificates.
Protected payloads stay local
Configurations, command output, diffs, backups, detailed inventory, query results and diagnostic evidence remain in your environment. The cloud sees workflow structure and sanitized run state.
The Edge dials out
No site needs an internet-facing WeftRoute port, and none accepts an inbound control-plane connection. Identity is bound to the enrolled Edge.
Local policy has the final say
A signed cloud instruction is not automatically safe. Your Edge revalidates capability, scope, local policy and the active change window before it touches anything.
Plans
Use the cloud where it helps. Keep control where it matters.
The choice is only ever we host it or you host it. Execution runs on your own setup either way, credentials never leave your environment either way, and the same flows run unchanged across both.
WeftRoute Cloud
Early accessYou host nothing. We run the control plane; you run only a small Edge at each site.
- Hosted, always-current control plane
- Multi-site coordination and scheduling
- Credentials and payloads stay at your Edge
- Outbound-only Edge link, no inbound ports
- Optional cloud AI authoring, off by default
Air-gapped
Ships todayEverything on your own server, inside your network. Control plane, Edge and evidence never leave the estate.
- No outbound connection of any kind
- Images, SBOMs, licence and model artifacts import as signed media
- Signatures and checksums re-verified inside the destination
- Offline answers for private CA, OIDC, DNS, NTP and backups
- Leave the network on if you want on-premise without full isolation — same appliance
Founding design partners
Help shape the platform your team will actually operate.
Bring one multi-site workflow and one non-negotiable trust requirement. We will map both into an early-access design session and build the flow with you.
Or email karmi.business@gmail.com directly.