LocalAI Self-hosted Citizen Developer

LocalAI Self-hosted

LocalAI Self-hosted Citizen Developer LLM Open Source OpenAI API Compatible Privacy GGUF Hugging Face CPU GPU Docker Model Selection Chatbot Automation

Platform	API Compatible	GPU Required	Cost	Privacy	เหมาะกับ
LocalAI	OpenAI	ไม่จำเป็น	Free	100%	Self-hosted
Ollama	Custom + OpenAI	ไม่จำเป็น	Free	100%	Developer
LM Studio	OpenAI	ไม่จำเป็น	Free	100%	Desktop
vLLM	OpenAI	จำเป็น	Free	100%	Production GPU
OpenAI API	Native	Cloud	Pay-per-use	ส่งออก	Best Quality

LocalAI Setup

=== LocalAI Installation ===

Docker — Quick Start

docker run -p 8080:8080 localai/localai:latest

Docker Compose — Production

version: '3.8'

services:

localai:

image: localai/localai:latest-aio-cpu

ports:

"8080:8080"

volumes:

./models:/build/models
./config:/build/config

environment:

THREADS=8
CONTEXT_SIZE=4096
DEBUG=false

restart: unless-stopped

Download Models

curl -L "https://huggingface.co/TheBloke/Mistral-7B-Instruct-v0.2-GGUF/resolve/main/mistral-7b-instruct-v0.2.Q4_K_M.gguf" \

-o models/mistral-7b.gguf

Model Config — models/mistral.yaml

name: mistral

backend: llama-cpp

parameters:

model: mistral-7b.gguf

temperature: 0.7

top_p: 0.9

top_k: 40

context_size: 4096

threads: 8

Test API — OpenAI Compatible

curl http://localhost:8080/v1/chat/completions \

-H "Content-Type: application/json" \

-d '{

"model": "mistral",

"messages": [{"role": "user", "content": "Hello!"}],

"temperature": 0.7

Python — Use OpenAI SDK

pip install openai

from openai import OpenAI

client = OpenAI(base_url="http://localhost:8080/v1", api_key="not-needed")

response = client.chat.completions.create(

model="mistral",

messages=[

{"role": "system", "content": "You are a helpful assistant."},

{"role": "user", "content": "อธิบาย Python list comprehension"},

temperature=0.7,

max_tokens=1000,

)

print(response.choices[0].message.content)

from dataclasses import dataclass

@dataclass

class ModelChoice:

model: str

size: str

ram_required: str

speed: str

quality: str

use_case: str

models = [

ModelChoice("Phi-3 Mini", "2.3GB", "4GB", "เร็วมาก", "ดี", "Chat Code"),

ModelChoice("Mistral 7B", "4.1GB", "8GB", "เร็ว", "ดีมาก", "General Purpose"),

ModelChoice("Llama 3 8B", "4.7GB", "8GB", "เร็ว", "ดีมาก", "General Purpose"),

ModelChoice("CodeLlama 7B", "4.1GB", "8GB", "เร็ว", "ดีมาก (Code)", "Code Generation"),

ModelChoice("Mixtral 8x7B", "26GB", "32GB", "ปานกลาง", "ดีเยี่ยม", "Complex Tasks"),

]

print("=== Model Choices ===")

for m in models:

print(f" [{m.model}] Size: {m.size} | RAM: {m.ram_required}")

print(f" Speed: {m.speed} | Quality: {m.quality}")

print(f" Use Case: {m.use_case}")

Citizen Developer Workflows

=== AI Workflows for Citizen Developers ===

Workflow 1: Document Summarizer

from openai import OpenAI

client = OpenAI(base_url="http://localhost:8080/v1", api_key="x")

def summarize_document(text, max_words=200):

response = client.chat.completions.create(

model="mistral",

messages=[

{"role": "system", "content": f"Summarize in {max_words} words. Thai language."},

{"role": "user", "content": text},

)

return response.choices[0].message.content

Workflow 2: Email Draft Generator

def draft_email(topic, tone="professional"):

response = client.chat.completions.create(

model="mistral",

messages=[

{"role": "system", "content": f"Draft a {tone} email in Thai."},

{"role": "user", "content": f"Topic: {topic}"},

)

return response.choices[0].message.content

Workflow 3: Data Analysis Helper

def analyze_data(data_description, question):

response = client.chat.completions.create(

model="mistral",

messages=[

{"role": "system", "content": "You are a data analyst. Give Python code."},

{"role": "user", "content": f"Data: {data_description}\nQuestion: {question}"},

)

return response.choices[0].message.content

@dataclass

class Workflow:

name: str

input_type: str

output: str

model: str

time: str

difficulty: str

workflows = [

Workflow("Document Summary", "Text/PDF", "สรุปย่อ", "Mistral 7B", "5-15 sec", "ง่าย"),

Workflow("Email Draft", "Topic + Tone", "Draft Email", "Mistral 7B", "5-10 sec", "ง่าย"),

Workflow("Code Generator", "Description", "Python Code", "CodeLlama 7B", "10-30 sec", "ง่าย"),

Workflow("Data Analysis", "Data + Question", "Analysis + Code", "Mistral 7B", "10-20 sec", "ปานกลาง"),

Workflow("Chatbot", "User Message", "AI Response", "Llama 3 8B", "3-10 sec", "ปานกลาง"),

Workflow("Translation", "Text + Language", "Translated Text", "Mistral 7B", "5-15 sec", "ง่าย"),

]

print("\n=== Citizen Developer Workflows ===")

for w in workflows:

print(f" [{w.name}] Difficulty: {w.difficulty}")

print(f" Input: {w.input_type} → Output: {w.output}")

print(f" Model: {w.model} | Time: {w.time}")

Production Considerations

# === Production LocalAI Setup ===

# Hardware Requirements
@dataclass
class HardwareReq:
    model_size: str
    ram: str
    cpu: str
    gpu: str
    users: str
    cost_estimate: str

requirements = [
    HardwareReq("3B (Phi-3)", "8GB", "4 cores", "ไม่จำเป็น", "1-5", "$500 PC"),
    HardwareReq("7B (Mistral)", "16GB", "8 cores", "แนะนำ", "5-20", "$1,000 PC"),
    HardwareReq("13B (Llama)", "32GB", "8+ cores", "แนะนำ", "10-50", "$2,000 PC"),
    HardwareReq("8x7B (Mixtral)", "64GB", "16 cores", "จำเป็น", "20-100", "GPU Server"),
]

print("Hardware Requirements:")
for h in requirements:
    print(f"  [{h.model_size}] RAM: {h.ram} | CPU: {h.cpu}")
    print(f"    GPU: {h.gpu} | Users: {h.users} | Cost: {h.cost_estimate}")

# Cost Comparison
comparison = {
    "LocalAI (Mistral 7B)": {"setup": "$1,000 one-time", "monthly": "$50 electricity", "per_1k_tokens": "$0"},
    "OpenAI GPT-4o": {"setup": "$0", "monthly": "Variable", "per_1k_tokens": "$0.005-$0.015"},
    "OpenAI GPT-3.5": {"setup": "$0", "monthly": "Variable", "per_1k_tokens": "$0.0005-$0.0015"},
}

print(f"\n\nCost Comparison (10K requests/month):")
for name, costs in comparison.items():
    print(f"  [{name}]")
    for k, v in costs.items():
        print(f"    {k}: {v}")

เคล็ดลับ

Start Small: เริ่มจาก Phi-3 Mini ทดลองก่อน แล้วค่อยขยาย
Q4_K_M: ใช้ Quantization Q4_K_M ดีทั้งคุณภาพและขนาด
OpenAI SDK: ใช้ OpenAI SDK เปลี่ยนแค่ base_url ง่าย
Privacy: เหมาะกับข้อมูลลับ ไม่ส่งออกนอกเครื่อง
Docker: ใช้ Docker Compose สำหรับ Production Setup

LocalAI คืออะไร

Open Source AI Server รันเครื่องตัวเอง OpenAI API Compatible LLM Image Audio Embeddings GGUF Hugging Face CPU GPU Docker Privacy Free

LocalAI Self-hosted Citizen Developer — รัน AI