Documentation

Akumi is an EU-sovereign, OpenAI-compatible inference platform. Send chat and embedding requests to a single base URL, layer governed modules like the PII firewall, guard, RAG, and memory onto any request, and get every call metered and audited. These docs cover the API and each service.

Your first request

The API is OpenAI-compatible, so point any client at the base URL with your key. Create one under API keys, then send a chat completion.

curl https://api.akumi.cloud/v1/chat/completions \
  -H "Authorization: Bearer $AKUMI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "mistral/mistral-small-2603", "messages": [{"role": "user", "content": "Hello"}]}'

Prefer typed access? Reach for a client SDK for PHP, TypeScript, or Python.

Explore the platform

Inference

Model router

Route across EU and external providers with aliases and fallback.

Embeddings

Generate embeddings with EU or external models through one endpoint.

Governance

PII firewall

Pseudonymize personal data before it reaches a model.

Guard

Block unsafe or off-policy content in the inference path.

Knowledge

RAG

Answer from your own documents with tenant-scoped retrieval.

Memory

Persist long-term facts and conversation threads.

MCP server

Connect Claude and other MCP clients to your governed knowledge.

Operations

Observability

Trace latency, cost, and pipeline stages per request.

Cache

Reuse stored responses to cut cost and latency.

Audit log

Review a metadata-only trail of platform activity.

Not sure where to begin? Create an API key, send the request above, then add a governed module such as the PII firewall or RAG. For typed clients, see the SDKs; for every endpoint, the API reference.