Keel is built for teams where security isn't a feature — it's the requirement. The log data plane runs entirely on your own infrastructure with no phone-home, and you bring the AI model you trust — a hosted key you approve, or a fully local model where nothing leaves at all. Here's exactly how your data stays yours.
Keel's guarantees come from how it's built — a self-contained appliance you run inside your own perimeter — not from a promise to handle your data carefully in someone else's cloud.
Keel is a single VM appliance you run inside your own environment (VMware, Hyper-V, KVM). All ingest, storage, search, and dashboards run on your hardware, and nothing goes to an Infranexis service — no phone-home, no telemetry.
Choose a local model and it can run fully air-gapped, with no outbound connectivity at all.
Keel ships no bundled model. You point it at the AI you choose — your own Anthropic or OpenAI key, any OpenAI-compatible endpoint, or a local model on your own network — with a primary and a failover backend. When AI runs, only the context for that one request goes to the endpoint you control; in local mode, nothing leaves the network at all.
If the model endpoint is down or you never configure one, Keel degrades to a complete conventional log platform.
Agents you connect over the MCP server are strictly read-only — they can search and investigate, but queries are SELECT-only, bounded, and scoped to the key, with no tools to change anything. The built-in assistant can take actions like adding a source or drafting a detection, but every change asks you to confirm first, is RBAC-scoped, and is audited — it never acts silently.
That contains the blast radius of prompt injection through log data: the worst case is a wrong answer or a proposed change you'd decline — never a system quietly altered. Humans keep the pager and the keyboard.
Single sign-on via LDAP/RADIUS, SAML, and OIDC, multi-factor authentication, role-based access control with teams, and audit logging of mutating actions — so access is governed by your existing identity provider and reviewable.
License verification is offline (RS256) — no outbound license calls, so it works in disconnected and air-gapped deployments. Updates are delivered as signed bundles you apply on your schedule.
Every deployment is your own dedicated instance — not a shared multi-tenant service. Run it as a single node, or as a 3-node HA cluster with a floating VIP for resilience.
The hardest part of adopting a new tool in a regulated environment is usually the vendor security review — because your data ends up in the vendor's cloud. With Keel, it doesn't.
Because Keel's log data plane runs entirely on your infrastructure, no Infranexis cloud or third-party service ever holds your logs, and there's no vendor data-store to include in a data-flow review. The one deliberate, controllable exception is the AI: when you trigger it, the context for that request goes to the model endpoint you chose — your own provider and key — and you can eliminate even that by running a local model, so nothing leaves at all. Your data stays governed by the same controls, retention rules, and audits you already run.
That's what makes Keel usable where data cannot leave the building: banking, healthcare, government, defence, and operational-technology (OT) networks — including strict data-residency and sovereignty requirements, and OT/ICS frameworks such as NERC CIP and IEC 62443.
Keel does not yet hold third-party attestations such as SOC 2 or ISO 27001 — and honestly, the appliance model is designed so you don't have to rely on one: your logs never reach us, so there's no vendor environment to certify.
Even with no SaaS environment to certify, we make it straightforward for your security team to sign off. During an evaluation we'll complete your security questionnaire (SIG, CAIQ, or your own), share architecture and data-flow documentation that shows exactly what does and doesn't leave your network, and walk your team through the design and our signed-release process.
On the roadmap: formal ISO 27001 and SOC 2 attestations as we scale — prioritised by what your procurement actually requires. Questions? [email protected].
It's the same appliance either way — only where it lives changes. Every option is single-tenant, and your data stays in your own environment.
Import the VM image (OVA / VHDX / QCOW2) into VMware, Hyper-V, or KVM. Runs fully on-premises and air-gap capable, with no outbound connectivity required.
The same QCOW2 image runs on private-cloud platforms like OpenStack and OpenShift Virtualization — single-tenant, in your own tenancy. Dedicated AWS / Azure / GCP marketplace images are on the roadmap; ask us if you need one for a specific account or sovereign region.
Prefer not to operate it yourself? Talk to us about a dedicated, single-tenant instance we deploy and run in your own environment, so your team stays hands-off while the data stays in your tenancy.
Whatever your data-handling, air-gap, or compliance constraints, the answer usually starts with "it runs in your environment." Let's walk through yours.
Book a walkthrough Try the live demo