offlineai-local 0.1.0

Offline local-LLM helper (llama.cpp + GGUF) with Python API and CLI.

offlineai-local

Small Python package to run a local GGUF model (via llama-server) with:

• Python API: get_answer(prompt, instruction="")
• CLI: offlineai ask ...
• Bootstrap and bundle helpers to avoid copying many loose files.

Publish to PyPI (Windows, D drive only)

1. Tao token tren https://pypi.org/manage/account/token/.
2. Mo PowerShell trong D:\AI_LOCAL_SYSTEM.
3. Gan token:

$env:PYPI_TOKEN = "pypi-xxxxxxxxxxxxxxxx"

4. Build + upload:

.\publish_to_pypi.ps1

Neu muon build wheel moi truoc khi upload:

.\publish_to_pypi.ps1 -BuildWheel

Jupyter localhost test from pip

Trong notebook:

%pip install offlineai-local

Khoi tao data local (1 lan):

from offlineai import bootstrap_from_zip
bootstrap_from_zip(r"D:\path\offlineai_bundle.zip")

Hoi:

from offlineai import get_answer
get_answer("Ban co the noi tieng Viet khong?")
get_answer(
    "Viet ham Python doc JSON an toan",
    "chi dua code python khong giai thich",
)

Quick start

1. Install package:

pip install .

2. Put runtime/model in one of these:

• %USERPROFILE%\.offlineai\runtime and %USERPROFILE%\.offlineai\models
• or bootstrap from a project folder / bundle zip

3. Check:

offlineai doctor

4. Ask:

offlineai ask "ban co the noi tieng viet khong"
offlineai ask "viet ham python doc json an toan" --code-only

Bootstrap from current project folder

If your project already has runtime/ and models/:

offlineai bootstrap-dir D:\AI_LOCAL_SYSTEM

Create a single portable bundle zip

offlineai pack --source D:\AI_LOCAL_SYSTEM --output D:\AI_LOCAL_SYSTEM\offlineai_bundle.zip

Then on a new machine:

pip install offlineai-local-0.1.0-py3-none-any.whl
offlineai bootstrap-zip D:\path\offlineai_bundle.zip
offlineai doctor

Notes

• This package does not train anything before each use.
• It calls local model directly (API-style).
• Model quality is high for size, but no model guarantees 100% correctness for all questions.