OPEN SOURCE · APACHE-2.0RUST · NO PYTHON IN THE CONTROL PLANE

The runtime between your application and any GPU.

OhMyGPU runs open models behind OpenAI-compatible APIs and owns the GPU runtime, model files and lifecycle. The same binary serves a laptop on 127.0.0.1 and a rented GPU box — your application just talks HTTP.

SHIPPING TODAY — V0.4.0

A headless local runtime on llama.cpp and whisper.cpp: macOS, Linux and Windows, OpenAI-compatible /v1 on 127.0.0.1 — text, tool calls, image input and speech to text — a verified catalog of 20 models plus any Hugging Face GGUF. The ten-minute GPU-box promise is the roadmap, stated with its budget — not fine print.

FRESH MACHINE — ANY PLATFORMbash
# install: two binaries, nothing else
$ omg serve
$ omg model pull qwen2.5-7b-instruct
 qwen2.5-7b-instruct 4.7 GB ▓▓▓▓▓▓▓▓░░ 82%
$ omg run qwen2.5-7b-instruct
 state: running · http://127.0.0.1:10692
$ curl localhost:10692/v1/responses \
    -d '{"model":"qwen2.5-7b-instruct",
         "input":"Why is the sky blue?"}'
0
Python in the control plane
2
Inference APIs, one pipeline
20
Models in the verified catalog
10:00
Minutes, model → endpoint · target
Architecture

The runtime is the product. Everything else is a client.

YOUR APPLICATION
Electron · Swift · Python · Tauri · .NET · Go — the OpenAI client it already has
HTTP · 127.0.0.1:10692
OHMYGPU RUNTIME
HEADLESS DAEMON
/v1/responses · /v1/chat/completions — inference, OpenAI-compatible subset
/v1/audio/transcriptions — speech to text
/ohmygpu/v1/* — model & runtime management
supervised subprocess per running model
llama.cpp
SHIPPED · AUTO-INSTALLED
whisper.cpp
SHIPPED · SPEECH TO TEXT
vLLM
PLANNED · STEP 4
SGLang
LATER
MLX
CONSIDERING
Candle
ARCHIVED
TensorRT-LLM
NOT PLANNED
METAL
CUDA
VULKAN
CPU
ComponentRole
ohmygpu-runtimeThe product. Inference APIs, Management API, model lifecycle, backend supervision.
ohmygpu · omgThe CLI — an administrative client of the runtime. Thin by design.
future GUIAn optional client. Nothing in the runtime depends on a UI.
third-party appsThe primary consumers. They talk to the runtime over HTTP.

OhMyGPU is not a chat app, a desktop app, an agent framework or a cloud platform. It is the boring, reliable local runtime underneath those things.

Your code says ensure model exists → start model, then uses either inference endpoint. Every state, download percentage and failure reason is visible over HTTP — nothing requires the CLI.

API reference
Scope

What your app no longer has to manage

CUDA · Metal
GPU / VRAM detection
model formats
downloads & storage
inference binaries
processes & ports
crashes & restarts
runtime configuration
Compatible
OpenAI-compatible, twice

/v1/responses is the canonical surface; /v1/chat/completions keeps the ecosystem working. Both reach the same model through one internal pipeline — verified against the official OpenAI SDK.

Explicit
A lifecycle you can see

not_installed → downloading → installed → starting → running — plus error with the reason and the tail of the engine log. States, progress and failures are all plain HTTP.

Data, not code
Recipes

Adapting a model is editing a YAML file — repo, quant variants, VRAM requirements, engine args, a smoke test — not writing code. Anyone can contribute one; the catalog grows in public.

Vision — where this goes

Any model, any GPU box, in ten minutes.

New open models ship every day; most never get a hosted API. Using them means hand-assembling GPU image, engine, weights, flags and proxy — differently for every model. The goal after v0.1: drop OhMyGPU onto any machine with a GPU, point it at a model — including one released today — and have an OpenAI-compatible endpoint inside ten minutes.

Stated as a budget, not a slogan: base install ≤ 1 min, backend pre-baked ≈ 0, weights bandwidth-bound with an honest ETA, warm-up ≤ 3 min — and once the nightly test matrix ships, the ten minutes get measured in public, not asserted.

1Remote-safe mode — bind + API key, CLI --hostPLANNED
2Linux packaging — install.sh, Docker, systemdPLANNED
3Multi-GPU detection + snapshot downloadsPLANNED
4vLLM backend — day-0 coverage for new modelsPLANNED
5Recipes as data — adapt models without codeIN PROGRESS
6Pre-baked CUDA image + nightly ten-minute matrixPLANNED
Catalog

Verified models, ready to pull

IDQuantSizeCapabilities
qwen2.5-7b-instructQ4_K_M4.7 GBTOOLS
llama-3.1-8b-instructQ4_K_M4.9 GBTOOLS
qwen3-4b-instructQ4_K_M2.5 GBTOOLS
qwen2.5-vl-7b-instructQ4_K_M4.7 GBVISION
whisper-large-v3-turboq8_00.87 GBSPEECH
…and any other GGUF via hf:owner/repo/file.gguf or a direct URL.All 20 models
News

Adaptation log

All news

Point it at a model.

Free, open source, Apache-2.0 — and staying that way.