# Akumi Platform: nine AI services on one EU-resident API One OpenAI-compatible API, served from the EU, with nine services you switch on per request and meter on their own. Recorded from the first token. ## Everything you need to run AI, and prove it. One OpenAI-compatible API, served from the EU, with nine services you switch on per request and meter on their own. ## Switch them on per request. Pay for what ran. Start with inference and routing, then add the firewall, Recall, cache and tracing as you grow. Every one of them writes to the same audit trail. ### Build Get a model answering, fast. - Inference API: A drop-in, OpenAI-compatible endpoint served from the EU. Change one base URL. - MCP Server: Memory and tools for AI clients, over the Model Context Protocol. - Cache: Serve a stored answer for a repeated request and skip the model. ### Govern Decide what a model may see and say. - Model Router: One endpoint for every model, with policy-based routing and a fail-closed egress guard. - Firewall: Pseudonymize personal data before any model sees it, and restore it in the response. - Recall: Your documents and your model's memory in one knowledge graph, with sources and permissions. ### Prove Show an auditor what happened. - Audit Trail: Every request from every product, recorded as metadata and retained on your policy. - Compliance Center: Art. 30 records and EU AI Act documentation, generated from what actually ran. - Observability: Trace every request stage by stage, with latency, cost and error rate. ## Not a setting you remember to turn on. ### EU-resident by default The application and your data stay in the EU, always. A model call leaves only when you have allowed it, and never without a record. ### Fail-closed egress The router refuses a non-EU model unless you have allowed it. The default is to stop, not to send. ### Metadata, never prompts The audit trail records what ran, where and under which policy. It never stores what was said. ## Everything included. Unlimited members. Every plan comes with all nine services and no per-seat charges. ### Pay as you go €0 to start - Credits at list price, €10 minimum top-up - 1 workspace · 14-day audit history ### Team €179 / month - 75,000 credits · 1 GB Recall storage - 3 workspaces · 90-day audit history ### Business €599 / month - 350,000 credits · 5 GB Recall storage - 15 workspaces · 12-month audit history · SSO and SCIM ### Enterprise Custom - Unlimited Recall storage and workspaces - 24-month audit history Every plan is EU-resident and audited. Services you do not switch on cost nothing. ## Questions people ask about the platform. ### Do I have to change my code? No. It is an OpenAI-compatible endpoint, so you change one base URL and keep your existing client. Chat, tools, streaming and structured output all work as they did. ### Where does inference actually run? On EU-resident models, unless you have allowed a specific model that is not. Anything else the router refuses. Every request records the region it actually ran in, so residency is something you can show rather than something you assert. ### Can I use a model outside the EU? Only if you allow it. The router is fail-closed, so a non-EU model is refused by default and the refusal is recorded like any other event. ### What does a service cost when it is off? Nothing. Each service is a per-request toggle metered on its own, so you pay for the ones that ran on the calls they ran on, and nothing for the rest. ### What is stored in the audit trail? Metadata only. What ran, when, in which region and under which policy. Prompt content is never stored, and retention follows your own policy. ## One platform. Every call on the record. Create a key, change one base URL, and switch on the services you need. The ones you leave off cost nothing. Registered in the Netherlands. No US parent.