hako
From-scratch C engine that runs hako-sho 3B and hako-koi 7B models on CPU.
Tested on Linux
Use it
Linux x86_64
download hakm.sh proxy script
curl -fsSL https://mithraeums.github.io/hakm.sh -o /tmp/hakm.shdownload actual install.sh
curl -fsSL https://raw.githubusercontent.com/mithraeums/hako/main/install.sh -o /tmp/install.shinstall hakm release binary via script
sh /tmp/install.shdownload tiny SmolLM2 Q8_0 GGUF model
mkdir -p ~/models && curl -fL -o ~/models/SmolLM2-135M-Instruct-Q8_0.gguf https://huggingface.co/bartowski/SmolLM2-135M-Instruct-GGUF/resolve/main/SmolLM2-135M-Instruct-Q8_0.ggufprepare replayable converter and model
curl -fsSL https://raw.githubusercontent.com/mithraeums/hako/main/tools/gguf2mlf.py -o ~/gguf2mlf.py && python3 ~/gguf2mlf.py ~/models/SmolLM2-135M-Instruct-Q8_0.gguf ~/models/smol135.mlf2
Check that it works
hakm --helpThe output should contain “the hako engine”.
First thing to try
hakm ~/models/smol135.mlf2 --infoRun hakm <model.mlf2> --info to check that a model file loads.
What it does
- Chat with a model over several turnsTested
How to use it
- Run
hakm <model.mlf2> --sys "<system prompt>" -n 16 -t 0to start a chat. - Type your message and press Enter to get a reply.
- Run
- Generate plain text completionsTested
How to use it
- Run
hakm <model.mlf2> --raw -t 0 -n 16 "<your text>"to continue the given text.
- Run
- Keep a model loaded and reuse its cacheTested
How to use it
- Run
hakm <model.mlf2> --serve -t 0 -n 16to keep the model loaded. - Send requests on standard input; replies start with
R.
- Run
- Run 3B and 7B models on the CPUNot tested
How far pickfit got
Linux x86_64
- Installs
- Starts
- 3 of 4 features work
- Works offline
macOS: the project says it runs there; pickfit hasn’t tried it.
Good for
- generate raw greedy text completion
- run one ChatML chat turn
- chat in interactive multi-turn REPL
An alternative to llama.cpp and ollama.
How pickfit tested it
Each run starts from a clean machine and follows the steps above. How pickfit verifies
Linux x86_64
Its main function works
- Replayed from scratch
- Works offline
- Checked on 1 Nov 2026
- Version v0.0.4
Differences the test machine hit
- README: curl -fsSL https://mithraeums.github.io/hakm.sh | sh
In the test: downloaded hakm.sh and install.sh, read both, then ran sh /tmp/install.sh; release tarball fetched at install time is a binary that cannot be pre-read - README: ./hakm "$M" --info arch banner
In the test: hakm --info banner prints to stderr, exit 0; check stderr for arch line - README: install step 5 runs gguf2mlf.py
In the test: nothing to change; a note: the box ran gguf2mlf.py, downloaded, without reading it first, so pickfit did not inspect it