Questions, answered.

e21 OS is a secure AI operating system, and it is not a single product you buy. It is what the e21 products become when they run together: Akora and Lynk as the desktop and mobile shells, CosmicMem as the memory layer, Sona as the voice interface, Teikn and Tachyon as the secure transport, and AiR as the gateway every model request passes through. Each product is useful on its own — start with one, and the OS is what you have once enough of them are in place.

Because it does what an operating system does: it provides the universal services other software would otherwise each solve badly on its own. A conventional OS handles identity, scheduling, memory, input and output, and isolation. An AI OS handles the same jobs for software that thinks — identity for agents and the humans they act for, routing to decide which model serves a request, context that survives the prompt window, secure transport between agents and devices, and tenant boundaries so one failure is not everyone’s. If a product does not provide those services to other software, it is an application, however good.

No, and it would be a strange way to buy an operating system. Nobody migrates everything on a Tuesday. Each product stands alone and is useful by itself, so you run one, then another, and at some point you notice you are running on it. That is also what makes the absence of lock-in checkable rather than promised — you can remove a component and see what happens. A platform that only works whole is a platform you cannot leave.

Three things that are structural rather than configurable. The records that tell you what happened — telemetry, metering, the audit chain, the usage ledger — are typed so that prompt content cannot enter them, with a test in our CI scanning every column of every table to keep it that way. Audit exists at every tier including the smallest account, rather than being an enterprise upsell. And an agent is a principal in its own right, carrying a delegation chain, so an action can be attributed to both the agent and the human who authorised it. Security you can turn off is a setting; these are properties.

Early, and we would rather say so here than have you discover it. Most of the suite is in Alpha, access runs through the Alpha programme rather than self-serve signup, and pricing is announced at general availability. We do not yet hold an external security certification, and we are not going to imply otherwise while a market rewards implying it — ask us directly and we will tell you exactly where that work has reached. What exists is running: AiR serves production traffic, and the platform layer is live behind an allowlist. This answer is reviewed as things move; last reviewed 2026-08-22.

Organisations that take data security and sovereignty seriously — particularly regulated industries such as finance, insurance, healthcare and government. e21 is headquartered in the United Kingdom, and the architecture is designed for any organisation that needs AI infrastructure to be auditable and under its own control. We also work with resellers, MSPs and telcos delivering secure AI infrastructure to their own customers. It is not for anyone who wants a chat interface, or who needs a compliance certificate in hand today.

Akora is the desktop shell — a private AI assistant built around privacy and safeguarding. It runs on your desktop, provides analytical support and task automation, and has a fully private mode where execution stays local and nothing leaves the device. Where you do want a cloud model, it routes through AiR with your sensitive context kept off the wire.

CosmicMem is the memory layer — persistent, secure context that survives the prompt window, so agents recall past interactions instead of starting from nothing each session. It is built into Akora and into the core OS as a securely hosted central option. Separation runs in layers rather than as a single tenant boundary: organisation, department, and down to the individual desktop, with management that scales from one person to an enterprise.

Teikn provides zero-trust network tunnelling with cloud-hosted WAF endpoints, so internal AI infrastructure is never exposed to the open web. AiR is the gateway your software talks to instead of each model provider directly — one API across providers, every request metered and priced, and every error attributed to the gateway, the provider or the client. Tachyon is encrypted messaging built for agent-to-agent collaboration, with delivery guarantees and an audit trail.

Sona is the voice interface — conversational AI that understands context and intent, for voice-first experiences either embedded in a workflow or standing alone. It is the newest of the products and, like most of the suite, access runs through the Alpha programme.

No — and not as a policy you would have to take on trust. The telemetry, metrics, audit log and usage ledger cannot hold prompt content because of how they are typed, and a test in our CI scans every column of every table for a canary string to keep it that way. The exceptions are enumerated rather than left to be discovered: response bodies are held for 24 hours so that retried requests are safe, and two content-reading features that have not shipped yet are opt-in and off by default. Where an organisation’s obligations require a record of what was asked and answered, that lives at the enterprise layer as an explicit choice, never in core. A tenant in strict zero-data-retention mode turns all of it off and routes only to providers that have attested they do not retain or train on content.

Single-digit milliseconds. On our measured runs, routing through AiR was marginally faster than going direct, because our cloud’s peering to the provider more than pays for the extra hop. You do not have to take that on trust: every response carries an x-air-overhead-ms header with the gateway’s own cost for that request, and the full benchmark protocol is published so you can reproduce it — or falsify it — against your own workload.

Yes, and the interesting part is what they deliberately cannot see. AiR produces an immutable, priced usage record for every request, so IT and security teams can answer what a key, a workspace or a project spent, and a tamper-evident audit log covers who did what. None of it can contain your prompts. Knowing what you spent and not being logged are usually a trade-off. Here they are the same system.

Nothing is held back on features. The commitment is that self-hosted runs the same build rather than a lagging fork, and that audit, SIEM export, SSO and role-based access are never sold as add-ons — they are there at every tier, including the smallest account. Self-hosting is a licensed deployment rather than a free download, and the sensible first step is a conversation about your requirements and a proof of concept, because the shape of the deployment depends on what you need to keep inside your own boundary.

It depends how much of e21 OS you run. Akora and Lynk install on the devices you already have. The backend products are containerised, and deploy either on-premise inside your own infrastructure or as a managed service, so you decide where your data sits. There is also a private cloud variant of e21 OS for organisations that cannot have certain content hosted anywhere but their own estate.

The honest answer, rather than the reassuring one: we run in the UK today, and each model’s processing region is disclosed. We do not say “your data stays in region X”, because inference happens at the model provider and the gateway holds response bodies for 24 hours in the node’s region to make retried requests safe. Multi-region deployment is roadmap, not shipped. When a tenant can land on a node in their own region, the stronger claim becomes true and we will make it then.

Pricing is announced at general availability. What we can tell you now is the shape rather than the rate: a platform fee set at 5%, charged transparently on top-ups rather than buried, and usage passed through at a published rate card. Billing opacity is the most common complaint about this category, so the commitment is that the fee is published and your credits are not quietly expired. Until GA, talk to us about your requirements and we will be straight with you about cost.

A conversation about what you are trying to do and a proof of concept. There is no self-serve enterprise route today, and pretending otherwise would waste your time and ours. If you want to be among the first to use the products as they ship, the Alpha programme is the route in.

It scales with where you start. A single user installing Akora is up and running the same day. A team adopting shared memory and routing gets a guided onboarding. Rolling out the full e21 OS across an enterprise comes with dedicated consulting, because at that point the work is as much about your existing infrastructure as it is about ours.

Both. We run AI strategy workshops, executive sessions and readiness assessments for organisations working out what they should be doing before they decide what to buy. That is deliberately separable from the products — the assessment is worth having even if you conclude that none of our software is the answer.

Ongoing support and advisory as your deployment scales, and a direct line to the team building the software rather than a ticket queue. During Alpha that line is unusually short, which is part of the point of the programme for both sides.

Because the claims are checkable. Every response from our gateway reports the gateway’s own latency cost so you can measure our overhead yourself. Our benchmark protocols are published so you can reproduce or falsify them. Our telemetry cannot hold your prompts by construction rather than by promise. And when something is not built, not certified or not decided, we say so — including on this page. If a vendor’s claims are shorter and more confident than ours, it is worth finding out whether their architecture is better or their marketing is.