offgrid 0.1.3

Python client for OffGrid LLM - Run AI models completely offline

OffGrid Python Client

Python client library for OffGrid LLM - Run AI models completely offline.

Installation

pip install offgrid

Quick Start

import offgrid

# Connect to server
client = offgrid.Client()  # localhost:11611

# Chat
response = client.chat("What is Python?")
print(response)

# List available models
models = client.list_models()
for m in models:
    print(f"- {m['id']}")

Full Usage

Chat

from offgrid import Client

client = Client()

# Basic chat (uses first available model)
response = client.chat("Explain quantum computing")
print(response)

# Specify model
response = client.chat("Hello!", model="Llama-3.2-3B-Instruct")

# With system prompt
response = client.chat(
    "Write a poem about AI",
    model="Llama-3.2-3B-Instruct",
    system="You are a creative poet.",
    temperature=0.9
)

# Streaming
for chunk in client.chat("Tell me a long story", stream=True):
    print(chunk, end="", flush=True)

# Full conversation
messages = [
    {"role": "system", "content": "You are a helpful assistant."},
    {"role": "user", "content": "Hello!"},
    {"role": "assistant", "content": "Hi there! How can I help?"},
    {"role": "user", "content": "What's the weather like?"}
]
response = client.chat(messages=messages)

API Key Authentication

from offgrid import Client

# With API key (if server has OFFGRID_API_KEY set)
client = Client(api_key="your-secret-key")

# All requests automatically include the Authorization header
response = client.chat("Hello!")

Session Management

# Sessions preserve conversation context on the server
sessions = client.sessions

# Create a new session
session = sessions.create("my-chat")

# List all sessions
for s in sessions.list():
    print(f"- {s['name']}: {len(s.get('messages', []))} messages")

# Chat with session (context is preserved)
response1 = sessions.chat_with_session("my-chat", "My name is Alice")
response2 = sessions.chat_with_session("my-chat", "What is my name?")
# response2 will correctly reference "Alice"

# Add messages manually
sessions.add_message("my-chat", "user", "Hello")
sessions.add_message("my-chat", "assistant", "Hi there!")

# Get session details
details = sessions.get("my-chat")
print(f"Messages: {len(details['messages'])}")

# Delete session
sessions.delete("my-chat")

Server Statistics

# Get comprehensive server statistics
stats = client.stats()

# Server info
print(f"Uptime: {stats['server']['uptime']}")
print(f"Version: {stats['server']['version']}")

# Inference metrics
print(f"Total requests: {stats['inference']['aggregate']['total_requests']}")
print(f"Total tokens: {stats['inference']['aggregate']['total_tokens']}")

# System resources
print(f"CPU usage: {stats['resources']['cpu_usage_percent']:.1f}%")
print(f"Memory: {stats['resources']['memory_used_mb']}MB")

# RAG status
if stats['rag']['enabled']:
    print(f"Documents: {stats['rag']['documents']}")

Model Management

# List installed models
for model in client.list_models():
    print(model['id'])

# Search for models
results = client.models.search("llama", ram=8)
for model in results:
    print(f"{model['id']} - {model['size_gb']}GB")

# Download a model
client.models.download(
    "bartowski/Llama-3.2-3B-Instruct-GGUF",
    "Llama-3.2-3B-Instruct-Q4_K_M.gguf",
    progress_callback=lambda pct, done, total: print(f"\r{pct:.1f}%", end="")
)

# Delete a model
client.models.delete("old-model")

# Import from USB
imported = client.models.import_usb("/media/usb")

# Export to USB
client.models.export_usb("Llama-3.2-3B-Instruct-Q4_K_M", "/media/usb")

Knowledge Base (RAG)

# Add documents
client.kb.add("notes.md")
client.kb.add("meeting", content="Meeting notes from today...")
client.kb.add_directory("./docs", extensions=[".md", ".txt", ".pdf"])

# List documents
for doc in client.kb.list():
    print(f"{doc['id']}: {doc['chunks']} chunks")

# Search
results = client.kb.search("project deadline")
for r in results:
    print(f"[{r['score']:.2f}] {r['content'][:100]}...")

# Chat with Knowledge Base context
response = client.chat(
    "What are the main action items from the meeting?",
    use_kb=True
)

# Remove documents
client.kb.remove("notes.md")
client.kb.clear()  # Remove all

AI Agents (New in v0.2.3)

Run autonomous agents that can use tools to complete tasks:

# Run an agent task
result = client.agent.run(
    "Calculate 127 * 48 + 356",
    model="llama3.2:3b"
)
print(result["result"])

# List available tools
tools = client.agent.tools()
for tool in tools:
    status = "✓" if tool["enabled"] else "✗"
    print(f"[{status}] {tool['name']}: {tool['description']}")

# Toggle tools on/off
client.agent.disable_tool("shell")  # Security: disable shell access
client.agent.enable_tool("calculator")

# Complex multi-step tasks
result = client.agent.run(
    "Read the VERSION file and tell me what version it is",
    model="llama3.2:3b",
    max_steps=5
)

Built-in Tools:

• calculator - Evaluate mathematical expressions
• current_time - Get current date/time
• read_file - Read file contents
• write_file - Write content to files
• list_files - List directory contents
• shell - Execute shell commands
• http_get - Make HTTP GET requests

MCP Integration (New in v0.2.3)

Connect external tools via Model Context Protocol:

# Add an MCP server
client.agent.mcp.add(
    "filesystem",
    "npx -y @modelcontextprotocol/server-filesystem /tmp"
)

# List configured servers
servers = client.agent.mcp.list()
for s in servers:
    print(f"{s['name']}: {s['url']}")

# Test a server connection
result = client.agent.mcp.test(url="npx -y @modelcontextprotocol/server-github")
print(f"Found {len(result.get('tools', []))} tools")

# Remove a server
client.agent.mcp.remove("filesystem")

LoRA Adapters (New in v0.2.3)

Manage LoRA adapters for fine-tuned models:

# List registered adapters
adapters = client.lora.list()
for a in adapters:
    print(f"{a['name']}: {a['path']}")

# Register a new adapter
client.lora.register(
    "coding-assistant",
    "/path/to/code-lora.gguf",
    scale=0.8
)

# Get adapter details
adapter = client.lora.get("coding-assistant")

# Remove an adapter
client.lora.remove("coding-assistant")

System Configuration (New in v0.2.3)

# Get server configuration and feature flags
config = client.config()
print(f"Version: {config['version']}")
print(f"Multi-user mode: {config['multi_user_mode']}")
print(f"Agent enabled: {config['features']['agent']}")

# Get real-time system stats
stats = client.system_stats()
print(f"CPU: {stats['cpu_percent']}%")
print(f"Memory: {stats['memory_percent']}%")

Embeddings

# Single text
embedding = client.embed("Hello world")
print(f"Dimensions: {len(embedding)}")

# Multiple texts
embeddings = client.embed(["Hello", "World", "AI"])

System Info

# Check server health
if client.health():
    print("Server is running")

# Get detailed info
info = client.info()
print(f"Uptime: {info['uptime']}")
print(f"CPU: {info['system']['cpu_percent']}%")
print(f"Memory: {info['system']['memory_percent']}%")

Configuration

from offgrid import Client

# Default: localhost:11611
client = Client()

# Custom server URL
client = Client(host="http://192.168.1.100:11611")

# Just hostname (auto-adds http://)
client = Client(host="192.168.1.100:11611")

# With API key authentication
client = Client(api_key="your-secret-key")

# Custom timeout (for slow models)
client = Client(timeout=600)  # 10 minutes

# Combined options
client = Client(
    host="http://192.168.1.100:11611",
    api_key="your-secret-key",
    timeout=300
)

Automatic Retry

The client automatically retries failed requests with exponential backoff:

• Up to 3 retry attempts
• Delays: 1s → 2s → 4s between retries
• Only retries on connection errors, not HTTP errors

# Retries are automatic
response = client.chat("Hello!")  # Will retry on transient failures

Error Handling

from offgrid import Client, OffGridError

client = Client()

try:
    response = client.chat("Hello")
except OffGridError as e:
    print(f"Error: {e.message}")
    if e.code:
        print(f"Code: {e.code}")

Requirements

• Python 3.8+
• OffGrid LLM server running (offgrid serve)
• No external dependencies (uses only stdlib)

Links

• OffGrid LLM - Main project
• API Reference
• Issue Tracker

License

MIT License

Custom server URL

client = Client(host="http://192.168.1.100:11611")

Just hostname (auto-adds http://)

client = Client(host="192.168.1.100:11611")

Custom timeout (for slow models)

client = Client(timeout=600) # 10 minutes

## Error Handling

```python
from offgrid import Client, OffGridError

client = Client()

try:
    response = client.chat("Hello")
except OffGridError as e:
    print(f"Error: {e.message}")
    if e.code:
        print(f"Code: {e.code}")

Requirements

• Python 3.8+
• OffGrid LLM server running (offgrid serve)
• No external dependencies (uses only stdlib)

Links

• OffGrid LLM - Main project
• API Reference
• Issue Tracker

License

MIT License