Comparison
Akumi vs Zylon
Zylon asks you to install and run a complete AI platform inside your own perimeter. Akumi asks you for a base_url. We run the EU-resident platform, keep it patched and scaled, and give you an OpenAI-compatible API you can be building against this afternoon, with residency you can prove per request rather than per deployment.
The same anxiety, solved from opposite ends.
Zylon hands you a complete AI platform to install inside your own perimeter, on-premise or in your own cloud. Akumi runs the EU-resident platform and hands you a base_url. That single choice cascades into everything else: who owns the capacity, the upgrades and the uptime, whether there is a start date, and whether the infrastructure costs the same in a quiet month. If your organization wants to operate its AI platform, they built for that. If it wants to use one, we did.
The two things being compared
What each one actually is.
Neither is a wrapper. The question is which side of the perimeter the platform sits on.
Akumi
A managed EU-resident AI platform behind one OpenAI-compatible endpoint. EU-resident models answer by default, with a firewall, a knowledge graph for retrieval and memory, a response cache and per-request observability on the same call. Self-serve signup, published usage-based pricing, and nothing for you to deploy, patch or scale.
Zylon
A private AI platform installed in your environment: on-premise, in a private cloud VPC, or fully air-gapped with, in their words, no internet connection required. Built by the team behind the open-source PrivateGPT project. They describe a complete stack rather than a wrapper, with built-in authentication, logging, rate limiting and observability, OpenAI-compatible endpoints, and integrations for n8n and LangChain.
Side by side
A platform to run, or a platform that runs.
| Dimension | Akumi | Zylon |
|---|---|---|
| Who operates it | We do. Serving, scaling, upgrades, patching and capacity are ours, and there is nothing for you to run. | You do, inside your own infrastructure, with their software and support. |
| Getting started | Self-serve signup and a first request in minutes, with no call and no installation. | States production-ready in under one week, which is fast for a deployed platform but is still a project with a start date. |
| Pricing transparency | Usage-based credits at published rates. Nothing runs, nothing costs, and you can price it before you talk to us. | States a fixed-cost subscription with no per-token pricing. Prices are not published. |
| Cost when idle | Nothing. You pay for requests, not for capacity sitting there overnight and at weekends. | A fixed subscription plus the infrastructure it runs on, whatever the usage. |
| Build surface | OpenAI-compatible API, plus first-party PHP, TypeScript and Python SDKs and an MCP server. | States OpenAI-compatible endpoints, with n8n and LangChain integrations. |
| Retrieval | Conversation memory and your documents share one knowledge graph. Each collection is its own partition, and every response carries the sources it used. | Document retrieval, deployed and operated inside your environment. |
- Who operates itAkumiWe do. Serving, scaling, upgrades, patching and capacity are ours, and there is nothing for you to run.ZylonYou do, inside your own infrastructure, with their software and support.
- Getting startedAkumiSelf-serve signup and a first request in minutes, with no call and no installation.ZylonStates production-ready in under one week, which is fast for a deployed platform but is still a project with a start date.
- Pricing transparencyAkumiUsage-based credits at published rates. Nothing runs, nothing costs, and you can price it before you talk to us.ZylonStates a fixed-cost subscription with no per-token pricing. Prices are not published.
- Cost when idleAkumiNothing. You pay for requests, not for capacity sitting there overnight and at weekends.ZylonA fixed subscription plus the infrastructure it runs on, whatever the usage.
- Build surfaceAkumiOpenAI-compatible API, plus first-party PHP, TypeScript and Python SDKs and an MCP server.ZylonStates OpenAI-compatible endpoints, with n8n and LangChain integrations.
- RetrievalAkumiConversation memory and your documents share one knowledge graph. Each collection is its own partition, and every response carries the sources it used.ZylonDocument retrieval, deployed and operated inside your environment.
Claims about Zylon are taken from zylon.ai and were last checked on 28 July 2026. Products move: if something here is out of date or unfair, tell us at contact@akumi.eu and we will correct it.
What you take on
Installed software is software you now operate.
A platform deployed inside your perimeter is yours from that moment on. Somebody sizes the infrastructure, somebody applies the upgrades, somebody is paged when it stops at eleven at night, and somebody re-tests it after each release. That is not a criticism of the software. It is what deploying any platform means, and it is a standing commitment of engineering attention that in most teams comes out of product work.
Akumi is the other trade. The interface you depend on is an HTTP endpoint rather than a running system. There is no version to be behind, no upgrade window to schedule, and no rota.
| Dimension | Akumi | Zylon |
|---|---|---|
| Who applies upgrades | We do, behind the endpoint, with no window to schedule. | You do, in your environment, on your release cadence. |
| Who carries the pager | Nobody on your side. | Somebody on your side. |
| Capacity planning | Ours. Spikes are absorbed without a purchase. | Yours, sized in advance for the busiest hour. |
- Who applies upgradesAkumiWe do, behind the endpoint, with no window to schedule.ZylonYou do, in your environment, on your release cadence.
- Who carries the pagerAkumiNobody on your side.ZylonSomebody on your side.
- Capacity planningAkumiOurs. Spikes are absorbed without a purchase.ZylonYours, sized in advance for the busiest hour.
The bill
Fixed cost is predictable. It is also fixed.
Zylon states a fixed-cost subscription with no per-token pricing and no usage limits, which is genuinely attractive at high, steady volume: heavy users stop watching the meter. The same shape is unforgiving at the other end. A pilot, a seasonal workload or a feature still building adoption pays the full amount, and so does the infrastructure underneath it.
Akumi meters per token and per request at published rates. Their prices are not published, so a like-for-like comparison needs a call. Ours can be worked out from the pricing page before you speak to anyone.
| Dimension | Akumi | Zylon |
|---|---|---|
| Pricing model | Usage-based credits at published rates, with pay as you go and no monthly minimum. | States a fixed-cost subscription with no per-token pricing and no usage limits. |
| Cost of a pilot | Roughly the tokens you spend evaluating it. | The subscription, plus the infrastructure it runs on. |
| Finding out the price | Published. Read it now. | Not published. Contact them. |
- Pricing modelAkumiUsage-based credits at published rates, with pay as you go and no monthly minimum.ZylonStates a fixed-cost subscription with no per-token pricing and no usage limits.
- Cost of a pilotAkumiRoughly the tokens you spend evaluating it.ZylonThe subscription, plus the infrastructure it runs on.
- Finding out the priceAkumiPublished. Read it now.ZylonNot published. Contact them.
Trying it
A week to production is still a project.
Zylon states production-ready in under one week, which is fast for a platform that has to be installed, and considerably faster than most enterprise software manages. It is still a project with a kickoff, an environment, people from both sides, and a decision made before any of it starts.
The gap that matters is not one week against another. It is that Akumi lets you find out whether the thing fits before committing anyone. The evaluation happens before the decision rather than after it.
| Dimension | Akumi | Zylon |
|---|---|---|
| Trying it | Self-serve, no call, no environment to prepare. | An engagement, scoped and scheduled. |
| Time to a first real answer | Minutes. | States under one week to production. |
| If it turns out not to fit | You spent an afternoon. | You spent a deployment. |
- Trying itAkumiSelf-serve, no call, no environment to prepare.ZylonAn engagement, scoped and scheduled.
- Time to a first real answerAkumiMinutes.ZylonStates under one week to production.
- If it turns out not to fitAkumiYou spent an afternoon.ZylonYou spent a deployment.
The honest answer
Who should choose which.
Choose Akumi if
- EU residency satisfies your requirement, and the data does not have to sit inside your own walls.
- You do not want to deploy, operate, patch or upgrade an AI platform, however good the software is.
- You want published, usage-based rates rather than a subscription negotiated per deployment.
- Your usage is variable, and paying for idle capacity is hard to justify.
- You want to evaluate this afternoon rather than schedule an installation.
Choose Zylon if
- Your requirement is that the software runs on infrastructure you control, air-gapped if necessary.
- A fixed subscription suits you better than a meter, because volume is high and steady.
- Your organization already operates platforms of this kind and one more is a known cost.
- Data may not leave your perimeter at all, which no managed platform can satisfy.
FAQ
Questions people ask.
- What is the main difference between Akumi and Zylon?
- Where the platform runs. Zylon is installed inside your own perimeter and operated by you. Akumi is a managed EU-resident platform reached over an OpenAI-compatible API, operated by us. Everything else, pricing shape, time to first request, who handles upgrades, follows from that.
- Can I run Akumi on-premise?
- No. Akumi is a managed EU platform. The data is EU-resident and residency is recorded per request in the audit trail, but the platform itself is not something you install. If your requirement is specifically that the software runs on hardware you own, that is a different category of product.
- Is Akumi EU-resident?
- Yes. The application, the data and the default models are EU-resident, and routing to an external or non-EU model is blocked at a guard that fails closed unless you explicitly allow it.
- Which is cheaper?
- It depends on volume and on how much idle capacity you would be paying for. Zylon states a fixed-cost subscription, which suits high steady usage. Akumi is usage-based at published rates, which suits variable or growing workloads. Their prices are not published, so a precise comparison needs a conversation with them.
- How quickly can I be running?
- On Akumi, minutes: signup, a payment method, a key and one base_url. Zylon states production-ready in under one week, which is quick for a deployed platform but begins with a scheduled engagement rather than a signup form.
- What happens to my documents?
- They are ingested into a knowledge graph, partitioned per workspace and per collection. The partition key is derived server-side from the authenticated caller, so a request can only reach partitions inside its own organization. Every answer carries the sources it used, and the audit trail stores metadata only.
Weighing other options?
Skip the installation.
Change one base_url and send a real request today. No deployment, no hardware, no week-long onboarding before you learn whether it fits.