llmash
Ollama-compatible LLM server and CLI that runs models locally on Windows and Linux.
Tested on Linux
Use it
Linux x86_64
You need first
- A GPU is optional
download installer script
curl -fsSL https://raw.githubusercontent.com/omgitsbase/llmash/main/install.sh -o /tmp/llmash-install.sh && wc -c /tmp/llmash-install.sh && echo OKinstall llmash via script
sh /tmp/llmash-install.shdownload tiny GGUF model
mkdir -p ~/models && curl -fL -o ~/models/stories260K.gguf https://huggingface.co/ggml-org/test-model-stories260K/resolve/main/stories260K-f32.gguf && ls -l ~/models/
Check that it works
llmash --versionThe output should contain “llmash version is 0.5.9”.
First thing to try
OLLAMA_MODELS=$HOME/models llmash serve --port 11434curl -fsSL http://127.0.0.1:11434/api/tags && echo && curl -fsSL http://127.0.0.1:11434/v1/modelsRun OLLAMA_MODELS=<models folder> llmash serve --port 11434 to start the server, then send requests to http://127.0.0.1:11434.
What it does
- Run an Ollama-compatible LLM serverTested
How to use it
- Start the server with
OLLAMA_MODELS=<models folder> llmash serve --port 11434. - List models with
curl -fsSL http://127.0.0.1:11434/api/tagsorcurl -fsSL http://127.0.0.1:11434/v1/models. - Chat by posting to
http://127.0.0.1:11434/v1/chat/completionswith a JSON body giving "model" and "messages".
- Start the server with
- Classify text without generating a replyTested
How to use it
- Start the server with
OLLAMA_MODELS=<models folder> llmash serve --port 11434. - Post to
http://127.0.0.1:11434/api/classifywith JSON giving "model", "state" (your text) and "questions", each with a "type" and "instructions". - Read the probabilities for each question under "answers" in the reply.
- Start the server with
- Assemble custom model buildsNot tested
- Remove model refusalsNot tested
How far pickfit got
Linux x86_64
- Installs
- Starts
- 2 of 4 features work
- Works offline
Windows: the project says it runs there; pickfit hasn’t tried it.
Good for
- remove model refusals without changing weights
- run the LLM server in console
- find and install a draft model for a model
- pull model builds from Ollama or Hugging Face
- check and set model context length up to 4x
An alternative to Ollama and vLLM.
How pickfit tested it
Each run starts from a clean machine and follows the steps above. How pickfit verifies
Linux x86_64
It produces model output
- Replayed from scratch
- Works offline
- Checked on 30 Oct 2026
- Version 0.5.9
Differences the test machine hit
- README: curl -fsSL https://raw.githubusercontent.com/omgitsbase/llmash/main/install.sh | sh
In the test: installer fetches prebuilt llmash tarball and llama.cpp runtime tarball as binaries during the run; they cannot be pre-read