ABOUT

Named after a guardian that never slept.

Talos was the bronze automaton Hephaestus forged to guard Crete. It circled the island three times a day, tireless and watchful, and it did the same job every time without being asked. That is a fair description of what infrastructure work should be, and an unfair description of what it usually is.

WHY IT EXISTS

It started as a runbook that stopped scaling.

Talos began with Frappe and ERPNext infrastructure managed the way most of it is managed: SSH, a folder of shell scripts, and a person who remembers which server does what. That works for three servers. At fifteen it becomes the job. At forty it becomes the reason nothing else gets done.

The specific thing that pushed it from scripts to a product was not scale, though. It was the realisation that none of it was reviewable. A destructive operation ran because one person decided it should, produced output into a terminal that got closed, and left no record anybody could query afterwards. Every part of Talos that looks like bureaucracy — the job model, the gates, the ledger — exists because of that.

It grew outward from Frappe because the same problem exists one layer down and one layer up: the Docker host underneath, the cloud account around it, the cluster next to it. The bet is that one control plane over all of them is worth more than four good tools that do not know about each other.

WHO IS BUILDING IT

An independent project, not a company.

Talos is not a registered company. There is no funding, no team page and no sales department, and this is the sort of thing that is much better said on the about page than discovered during procurement.

Practically, it means three things. Nobody can take your money, which is why the software is free and self-hosted rather than free-for-now. There is no support contract, only a person answering email. And the licence question is open, which is listed below as something to fix rather than something to gloss.

WHERE IT ACTUALLY IS

What is proven, and what is not.

Feature lists flatten everything to the same confidence. This does not.

RUNNING IN ANGER

Servers, discovery and the job model
Frappe benches, sites and apps
Docker stacks and registries
Pipelines, approval gates and the audit ledger
Monitoring, alerts and live log streaming

BUILT AND TESTED, LESS MILEAGE

Databases, point-in-time recovery and restore verification
Kubernetes clusters and workloads
Deployments
AWS, Azure, GCP and Oracle managers

BETA — BUILT, NOT YET PROVEN

BETA
Hypervisor (Proxmox and libvirt) — never run against a live host
The AI Engine — never run end-to-end against a live provider key

WHAT IS NEXT

Four things, in roughly this order.

01

Prove the beta capabilities

Hypervisor against real Proxmox and libvirt hosts; the AI Engine end-to-end with a live key. Both lose their beta label by being exercised, not by being declared ready.

02

An independent security audit

Not scheduled. It should be, and saying so here is the honest version of a roadmap item nobody has funded yet.

03

Publish the licence

There is no LICENSE file in the repository yet. Until there is, the terms under which you can run this are genuinely unclear, which is not a good state for a self-hosted product.

04

Documentation depth

The docs cover installing and operating. Per-capability reference is not written.

On the name

There are other things called Talos in adjacent territory — Talos Linux from Sidero, and Cisco Talos in threat intelligence. Different categories, no affiliation with either. If you arrived here looking for one of those, that is our fault rather than yours.

It is free, and it runs on a machine you already have.

The fastest way to judge any of the above is to install it and point it at one server.

Install itGet in touch →