About

The specialty inference house for the models hyperscalers skip.

Tomoul is fourteen engineers building one Zig engine and one OpenAI-compatible cloud. The wedge is open-weights inference — embeddings, rerankers, speech and African-language LLMs — served from regions that matter to the people deploying them.

01 · Wedge

The models hyperscalers under-serve.

We host the open-weights checkpoints that AWS Bedrock and Azure don't have time for. Yoruba TTS. Swahili BGE. Igbo instruct. The long tail of capability that real products need.

42 models live · 11 languages
02 · Substrate

One engine. Hand-tuned, Zig.

Cloud, CLI and the open-source engine artifacts ship from the same source tree. When a kernel matters, we drop a layer and write it. The repo is at github.com/tomoul/tomoul.

84-line fused kernel · 4.2× RTF
03 · Geography

EU today. Africa as it earns its rack.

Helsinki and Frankfurt serve traffic now. Lagos in Q3, Nairobi in Q4, Joburg and Cairo as demand validates. Regions follow customers, not press releases.

2 regions live · 3 in build
The case

Open-weights infrastructure is still under-built.

The hyperscalers will host the same six closed models everyone else does, on the schedule that suits their largest enterprise customers. That leaves a long tail of useful open checkpoints — most of the multilingual world, most of the specialty workloads — running on a developer's M2 because no one bothered to make them a real product.

Tomoul is the company that bothered. We pick the models, we tune the kernels, we publish the prices, and we put them behind an OpenAI-compatible URL so the cost of trying us is one line of code.

The Africa bit

We are not "the African AI cloud."

Africa is a capability, not an identity. Concretely: payment rails for ten currencies, residency in Lagos and Nairobi as we earn the demand, eleven African-language model families on the menu, a footer that flips to four languages, and a team that lives across Helsinki, Lagos, Cairo and Cape Town.

Cassava builds the African AI cloud. Good for them. We build a clean, focused inference substrate that takes Africa seriously and is equally easy to use from Berlin.

What we are
  • An OpenAI-compatible inference cloud
  • An on-ramp for builders running open weights
  • The home for African-language LLMs and speech
  • A Zig engine and a small team that ships it
  • EU-resident today, Africa-resident soon
What we are not
  • An AWS killer
  • The first African AI anything
  • An open-source AWS / Cloudflare clone
  • A wrapper that resells someone else's API
  • A model lab — we host, we don't pre-train

The team.

Fourteen people across four offices. Half of us have shipped infrastructure at hyperscale; half of us are first-time founders. We are hiring — see open roles.

AD
Aïssatou Diallo
Co-founder · Engine
Wrote the first kernel. Owns the Zig engine and the kernels that make it fast.
prev: Modular · CERN
YE
Yara El-Sayed
Co-founder · Cloud
Designs the OpenAI-compatible surface and the console. Holds the pricing page.
prev: Vercel · AWS Lambda
TO
Tunde Okafor
Co-founder · Models
Picks the catalog. Runs the African-language model program. PhD in NLP from Edinburgh.
prev: Cohere · DeepMind
MP
Mateus Pereira
Co-founder · Ops
Brings up regions. Owns reliability. Will tell you why Hetzner is fine.
prev: Fly.io · Equinix

Backed by people we'd want to work with anyway.

Initialized Capital
Seed lead
Future Africa
Seed
Latitude Ventures
Seed
Operator angels
21 individuals
YC W26
Pre-seed
Working with us

Either build on top, or come build it.

If you have a product that needs the kind of inference we host, sign up at the console — it takes a minute. If you'd rather come help us build the engine, we have open roles in Helsinki, Lagos and remote.