The EU AI Act arrives August 2. Can you prove where your AI data goes?
Comparison

Akumi vs Aimable

Both of us are European, both put a governance layer in front of the model, and both expose an OpenAI-compatible API. The difference is what actually answers the request. Aimable redacts your data and forwards it to the frontier model you pick, most of which are American. Akumi serves the request on EU-sovereign models by default and blocks anything leaving the EU at a fail-closed guard.

The short answer

Both platforms sit between your sensitive data and a large language model, and both are European companies with a governance story. The decision comes down to one question: do you want your data redacted before it is sent to an American frontier model, or do you want it answered inside the EU in the first place?

Aimable tokenizes personal data and forwards the request to the model you select, including GPT, Claude and Gemini. Akumi serves the request on EU-sovereign models by default, and a request only reaches an external or non-EU model if you have explicitly allowed it, enforced by a guard that fails closed. If your compliance objection is that the data reaches a US provider at all, rather than that it reaches one unredacted, that single difference decides the evaluation.

What Akumi is

An EU-sovereign AI platform you build on. One OpenAI-compatible endpoint fronts EU-resident models, with the governance modules already wired in: a PII firewall, a guard that scans prompts and tool output, a knowledge graph for retrieval and memory, a response cache, and per-request observability. Self-serve signup, published pricing, and first-party SDKs for PHP, TypeScript and Python.

What Aimable is

An AI productivity suite for organisations with sensitive data, built around Spaces, Collections and Policies that end users work in directly. It detects and tokenizes personal data, then routes to the model you choose across GPT, Claude, Llama, Gemini and Mistral. It also exposes an OpenAI-compatible API and MCP. Available as an EU-hosted single-tenant cloud or deployed in your own AWS, Azure, GCP or on-premise environment.

Redact and forward, or keep it in the EU.

Dimension Akumi Aimable
What answers the request EU-sovereign models by default. Routing to an external or non-EU model is blocked at a fail-closed guard unless you explicitly allow it. The frontier model you select, including GPT, Claude and Gemini, with personal data tokenized before it is sent.
Getting started Self-serve signup, published pricing, and a first request in minutes without talking to anyone. Demo-gated. Every call to action books a demo or contacts sales, and pricing is quoted per deployment.
Time to production Minutes. A key and one base_url, with nothing to schedule. States 2 to 6 weeks to production, on either the cloud or the self-hosted path.
Governed retrieval Conversation memory and your documents share one knowledge graph, so an answer follows the link from a user to their organization to the document that answers them. Each collection is its own graph partition, namespaced to your organization and workspace before the query is built, and every response carries the sources it used. Curated Collections of documents, team-managed and kept current, searched per Space.
Personal data PII firewall on by default: pseudonymized before egress, restored in your response. Comparable: personal data is tokenized before the model and restored on the way back.
Developer surface OpenAI-compatible API, first-party PHP, TypeScript and Python SDKs, an MCP server, and a published OpenAPI spec. OpenAI-compatible API, MCP, space-scoped keys, and public developer docs.

Claims about Aimable are taken from aimable.ai and were last checked on 29 July 2026. Products move: if something here is out of date or unfair, tell us at [email protected] and we will correct it.

Redaction reduces exposure. It does not move the processing.

Tokenizing personal data before a prompt leaves your building is a genuinely good control, and Akumi does it too: the PII firewall is on by default and restores the original values in your response. What it does not do is change where the inference happens. Under GDPR, pseudonymized data is still personal data, so sending it to a US provider is still a transfer that needs a lawful basis, a DPA and standard contractual clauses. The firewall lowers the blast radius; it does not remove the paperwork.

That is why Akumi treats the model itself as the control point rather than only the payload. EU-sovereign models answer by default, and reaching anything outside that set is blocked unless you have explicitly enabled it and, where personal data is involved, recorded an acknowledgment. The guard fails closed, so a misconfiguration keeps data in rather than letting it out.

Dimension Akumi Aimable
Default destination An EU-sovereign model, with no configuration required to get there. The frontier model you select for the Space, including US-hosted options.
Leaving the EU Blocked at a guard that fails closed, and only ever opened by an explicit, recorded decision. A routing choice, with personal data tokenized before the request is sent.
If the config is wrong The request is refused. The failure mode keeps data inside the EU. The request proceeds to the configured model.

One knowledge graph, not a folder of documents.

Most retrieval is a document search: chunk the files, embed them, return the closest passages. That answers questions whose answer sits inside one document, and struggles with the ones that do not. Akumi keeps conversation memory and ingested documents in the same governed knowledge graph, so a question can follow the link from an end-user to their organization to the document that actually answers them, rather than hoping one chunk happens to contain the whole chain.

Isolation is the part worth reading closely. Each collection is its own partition of the graph, and the partition key is built server-side from the authenticated caller before the query is assembled. The graph engine never receives a tenant identifier it could be talked into changing: it only sees an opaque partition it was handed. A request cannot address another workspace or another customer, because the code never gives it the chance to ask.

Dimension Akumi Aimable
Shape of the knowledge Memory and documents in one graph, so answers can traverse the relationships between them. Curated Collections of documents, team-managed and kept current.
Isolation boundary Computed in the reviewed application layer from the authenticated caller, before the query exists. Organized around a curator who owns each Collection, scoped per Space.
Showing your work Every response carries a sources array naming the documents and facts it used. Cites sources in its answers.

You can evaluate this one without booking anything.

Aimable is sold the way enterprise software usually is: every call to action books a demo, pricing is quoted against your deployment, and their site states 2 to 6 weeks to production on either the cloud or the self-hosted path. For a suite being rolled out to a whole department, that is a reasonable shape.

It is a poor shape for an engineer with an afternoon and a question. Akumi publishes its prices, opens signup to anyone with a payment method, and answers the first real request in minutes against the same OpenAI-compatible endpoint you already write against. You find out whether it fits before anyone has to be persuaded internally that it might.

Dimension Akumi Aimable
Trying it Self-serve signup, no call, no scheduled onboarding. Demo-gated. Every call to action books a demo or contacts sales.
Knowing the price Published per plan and per meter, before you talk to anyone. Quoted per deployment.
Reaching production Minutes: a key and one base_url. States 2 to 6 weeks.

Why teams pick Akumi when the model matters.

Choose Akumi if

  • Your objection is that the data reaches a US provider at all, not merely that it reaches one unredacted.
  • You want EU-sovereign models serving the request by default, with non-EU routing off unless you turn it on.
  • You want to evaluate today, self-serve, against published prices and without a sales call.
  • You need per-request residency your own auditors can read off a log.
  • You are building a product rather than rolling out a tool for staff to log into.

Questions people ask

What is the main difference between Akumi and Aimable?

Where the request is answered. Aimable tokenizes personal data and forwards the request to the frontier model you choose, which is often US-hosted. Akumi serves it on EU-sovereign models by default and blocks non-EU routing at a guard that fails closed unless you explicitly allow it. Both redact personal data; only one keeps the inference in the EU as the default.

Is Aimable EU-hosted?

Yes. Aimable states an EU-hosted, single-tenant cloud, and also offers deployment into your own AWS, Azure, GCP or on-premise environment. The distinction Akumi draws is not about where the platform runs but about which model answers the request: an EU-hosted control layer can still forward the prompt to a US frontier model.

Does Akumi have an OpenAI-compatible API?

Yes. Akumi is OpenAI-compatible, so an existing integration moves by changing one base_url and the key. There are also first-party SDKs for PHP, TypeScript and Python, and an MCP server. Aimable also documents an OpenAI-compatible API and MCP, so neither platform requires you to rewrite your client.

Which models can I use on Akumi?

A curated roster centred on EU-sovereign models, with external providers available under an explicit routing policy. Models sit behind one endpoint, so switching is a string change rather than a new integration, and fallback is handled by the router.

How does Akumi keep one team's documents out of another team's answers?

Each collection is a separate partition of the knowledge graph. The partition key is derived server-side from the authenticated caller before the query is built, so a request can only ever address partitions inside its own organization and workspace. The isolation is enforced in code rather than by configuration or convention.

Can I prove where a specific request was processed?

Yes. Every request records the model, the provider and the region that served it in a metadata-only audit trail, so residency is answered per request from a log rather than from a policy document. Prompts and completions are never stored in that trail.

Do I need to talk to sales to try Akumi?

No. Signup is self-serve and pricing is published. You need a payment method before running inference, and then a key and one base_url gets you a first response in minutes.

How is Akumi priced?

Published plans with a monthly credit allowance, plus pay as you go with no monthly minimum. Base inference is metered per token and request, and each governance module meters on its own line when it runs, so you can see what the PII firewall, retrieval, guard and cache each cost.

Is Akumi GDPR compliant?

The platform is built around GDPR controls: EU residency, pseudonymization before egress, one-call erasure of an end-user's data, and metadata-only records. A DPA governs Akumi's role as your processor, and the sub-processor list is published. Compliance is a property of your whole processing activity, not only of a vendor, which is why the audit trail is designed to be evidence you can hand over.

Does Akumi support DORA and NIS2?

Yes, in the sense that matters for a supplier: the sub-processor list, the DPA and per-request residency records give a DORA register of information its entries and the evidence behind them, and the documented isolation, access control and encryption let a NIS2 entity evidence Akumi in its supply chain. ISO 27001 is in progress and not yet certified, and the trust page says so plainly.

Weighing other options?

Akumi vs Azure OpenAI Akumi vs calling the provider yourself Akumi vs self-hosting open models Akumi vs Zylon

Keep it in the EU, and start today.

Change one base_url and send a real request against your own documents, served on an EU-sovereign model. No demo, no deployment, no procurement cycle.