Skip to content
In development

One API for every model you use, governed from Ghana.

The Dekka AI Gateway gives your applications one OpenAI-compatible endpoint for OpenAI and Anthropic models. You bring your own keys, set a budget for each project and decide which providers a prompt may reach.

Example requestPOST /v1/chat/completions
What your application sends

Summarise this complaint from Kofi Mensah, who called from 024 412 7781 with Ghana Card GHA-713200941-6, about a mobile money transfer that never arrived.

  • anthropic/claude-sonnet-5-5Anthropic, United StatesPrimary
  • openai/gpt-4.1-miniOpenAI, United StatesFallback
  • dekka/privateYour private model in GhanaComing soon
Response, streamed

Point the OpenAI SDK at Dekka and use a Dekka key. Streaming works as it does today, and moving a feature to another provider means changing the model name.

import OpenAI from "openai"; const client = new OpenAI({  baseURL: "https://api.dekka.cloud/v1/ai",  apiKey: process.env.DEKKA_AI_KEY,}); const stream = await client.chat.completions.create({  model: "anthropic/claude-sonnet-5-5",  messages: [{ role: "user", content: "Summarise this complaint." }],  stream: true,});

Your own keys
Add your OpenAI and Anthropic keys once. Dekka encrypts them, never shows them again and uses them only for your projects.
A budget for every project
Set a monthly limit for each project, and the gateway refuses new requests when a project reaches it instead of running up a bill.
Fallback in the order you choose
When the first model on your list is down or rate-limited, the request moves to the next one without any change to your code.
Only the providers you allow
Each project lists the providers and models it may use. A request that would break those rules is refused rather than quietly sent somewhere else.
Usage you can read
Requests, tokens, latency and cost for every project and model, in the dashboard and through the API.
In development

Veil removes personal details before a prompt leaves.

Veil is a high-performance, locally executable, Ghana-aware personally identifiable information detection and redaction engine.

It will run inside the gateway in Ghana and recognise Ghana Card numbers, Ghanaian phone and mobile money numbers, names and addresses, so the model abroad sees placeholders where those details were.

Coming soon

Private local models are coming soon.

Open models running on GPUs in Ghana, behind the same endpoint and the same key. When a prompt can’t leave the country at all, you’ll send it to a private model and nothing else in your code changes.

model: "dekka/private"

Where your prompts go

The gateway runs in Ghana. The provider you choose decides where the model itself runs.

Where each model provider processes requests
ModelsProcessed in
Anthropic modelsUnited States
OpenAI modelsUnited States
Private local modelsGhana

Questions

Can I use the AI Gateway today?

It’s in development. Pilot teams get access first, so apply for the pilot to hear as soon as it opens.

Which models can I use?

OpenAI and Anthropic models first. Private open models running on GPUs in Ghana are coming soon, behind the same endpoint.

Do I bring my own provider keys?

Yes. You add your OpenAI and Anthropic keys, and each provider bills you for what you use, as it does today.

Does my data stay in Ghana?

The gateway runs in Ghana, but OpenAI and Anthropic process requests in the United States, so a prompt sent to them leaves the country. Veil will remove personal details before that happens, and private local models will keep prompts in Ghana entirely.

Be first to route your AI through Ghana.

Apply for the pilot