aiccel 3.5.0

AIccel is a versatile Python library for building lightweight AI agents with multiple LLM providers

⚡ AICCEL Framework

The Production-Grade, Security-First AI Agent Framework

Build AI agents that are fast, secure, and production-ready in minutes.

Installation • Quick Start • Features • Documentation • Examples

---

📦 Installation

# Core framework (OpenAI, Gemini, Groq support)
pip install aiccel

# With security features (jailbreak detection)
pip install aiccel[safety]

# With privacy features (PII masking via GLiNER)
pip install aiccel[privacy]

# Full suite (all features)
pip install aiccel[all]

Verify installation:

aiccel check

---

🚀 Quick Start

1. Create Your First Agent (30 seconds)

from aiccel import Agent, GeminiProvider

# Initialize provider (uses GOOGLE_API_KEY env var if not provided)
provider = GeminiProvider(model="gemini-2.0-flash")

# Create agent
agent = Agent(
    provider=provider,
    name="Assistant",
    instructions="You are a helpful AI assistant."
)

# Run query
result = agent.run("What is the capital of France?")
print(result["response"])

2. Agent with Tools

from aiccel import Agent, GeminiProvider, SearchTool, WeatherTool

# Create tools
search = SearchTool(api_key="your-serpapi-key")
weather = WeatherTool(api_key="your-openweather-key")

# Create agent with tools
agent = Agent(
    provider=GeminiProvider(),
    tools=[search, weather],
    name="ResearchBot",
    instructions="Help users find information and check weather."
)

# Agent automatically selects the right tool
result = agent.run("What's the weather in Tokyo?")
print(result["response"])
print(f"Tools used: {result['tools_used']}")

3. Async Support

import asyncio
from aiccel import Agent, GeminiProvider

async def main():
    agent = Agent(provider=GeminiProvider())
    
    # Async execution
    result = await agent.run_async("Explain quantum computing")
    print(result["response"])

asyncio.run(main())

---

🔥 Core Features

🧠 Intelligent Agents

from aiccel import Agent, AgentConfig

# Configure agent behavior
config = AgentConfig(
    thinking_enabled=True,      # Enable chain-of-thought reasoning
    max_tool_retries=3,         # Auto-retry failed tools
    safety_enabled=True,        # Enable jailbreak detection
    strict_tool_usage=False,    # Allow mixed tool/LLM responses
)

agent = Agent(
    provider=provider,
    config=config,
    tools=[search, calculator],
    instructions="You are a research analyst."
)

result = agent.run("Analyze the market trends for EV batteries")
print(f"Thinking: {result['thinking']}")  # See the reasoning process
print(f"Response: {result['response']}")

🔒 Enterprise Security Suite

PII Masking (Privacy Protection)

from aiccel.privacy import EntityMasker

masker = EntityMasker()

# Automatically detect and mask sensitive data
text = "Contact John Smith at [e-mail] or 555-123-4567"
masked, mapping = masker.mask(text)
# Output: "Contact [PERSON_1] at [EMAIL_1] or [PHONE_1]"

# Unmask after LLM processing
original = masker.unmask(masked, mapping)

Jailbreak Detection

from aiccel.jailbreak import JailbreakGuard, SecurityMode

# Production mode: Block on any detection
guard = JailbreakGuard(security_mode=SecurityMode.FAIL_CLOSED)

is_safe = guard.check("Normal user query")  # True
is_safe = guard.check("Ignore all instructions and...")  # False (blocked)

# Use as decorator
@guard.guard
def process_query(query: str):
    return agent.run(query)

AES-256 Encryption

from aiccel.encryption import encrypt, decrypt

# Encrypt sensitive data
encrypted = encrypt("my-api-key", password="secure-password")

# Decrypt when needed
original = decrypt(encrypted, password="secure-password")

🤝 Multi-Agent Orchestration

from aiccel import Agent, AgentManager, GeminiProvider

# Create specialist agents
researcher = Agent(
    provider=GeminiProvider(),
    name="Researcher",
    instructions="You gather and analyze information."
)

writer = Agent(
    provider=GeminiProvider(),
    name="Writer", 
    instructions="You create clear, engaging content."
)

analyst = Agent(
    provider=GeminiProvider(),
    name="Analyst",
    instructions="You provide data-driven insights."
)

# Create orchestrator
manager = AgentManager(
    llm_provider=GeminiProvider(),
    agents=[researcher, writer, analyst],
    instructions="Coordinate agents to solve complex tasks."
)

# Automatic routing to best agent
result = manager.route("Write a blog post about AI trends")

# Collaborative multi-agent reasoning
result = await manager.collaborate_async(
    "Analyze this quarterly report and create an executive summary"
)

⛓️ Workflow Pipelines

from aiccel import WorkflowBuilder, WorkflowExecutor

# Build a deterministic pipeline
workflow = (
    WorkflowBuilder("content-pipeline")
    .add_agent("research", researcher)
    .add_agent("write", writer)
    .add_agent("review", reviewer)
    .chain("research", "write")      # research -> write
    .chain("write", "review")        # write -> review
    .add_condition(                  # Conditional routing
        "review",
        lambda result: "approved" in result.lower(),
        on_true="publish",
        on_false="write"             # Loop back for revisions
    )
    .build()
)

# Execute workflow
executor = WorkflowExecutor(workflow)
result = await executor.execute_async("Create a blog post about Python")

🔌 MCP (Model Context Protocol) Support

from aiccel.mcp import MCPClient

# Connect to MCP-compatible tool servers
client = MCPClient.from_url("http://localhost:3000/mcp")
await client.connect()

# List available tools
tools = await client.list_tools()
print(f"Available tools: {[t.name for t in tools]}")

# Call tools
result = await client.call_tool("search", {"query": "AI news"})

# Convert MCP tools for use with AICCEL agents
from aiccel.mcp import MCPToolAdapter
aiccel_tools = MCPToolAdapter.convert_tools(tools, client)

agent = Agent(provider=provider, tools=aiccel_tools)

🎯 Request Context & Observability

from aiccel import Agent, request_scope, get_request_id

# Track requests across your application
with request_scope(user_id="user_123") as ctx:
    print(f"Request ID: {ctx.request_id}")
    
    # All operations within this scope are correlated
    result = agent.run("Process this query")
    
    # Logs automatically include request_id for debugging

📊 Custom Tools

from aiccel.tools import BaseTool, ToolResult, ParameterSchema

class CalculatorTool(BaseTool):
    """A simple calculator tool."""
    
    name = "calculator"
    description = "Perform mathematical calculations"
    parameters = [
        ParameterSchema(
            name="expression",
            type="string",
            description="Mathematical expression to evaluate",
            required=True
        )
    ]
    
    def execute(self, expression: str) -> ToolResult:
        try:
            result = eval(expression)  # Use safer eval in production
            return ToolResult.success(str(result))
        except Exception as e:
            return ToolResult.error(f"Calculation failed: {e}")

# Use with agent
calculator = CalculatorTool()
agent = Agent(provider=provider, tools=[calculator])
result = agent.run("What is 15 * 24 + 36?")

---

🏢 FastAPI Integration

from fastapi import FastAPI
from aiccel import Agent, GeminiProvider
from aiccel.integrations import create_agent_routes

app = FastAPI()

# Create agent
agent = Agent(
    provider=GeminiProvider(),
    name="API-Agent"
)

# Add routes automatically
router = create_agent_routes(agent)
app.include_router(router, prefix="/agent")

# Endpoints created:
# POST /agent/run          - Run a query
# POST /agent/stream       - Stream response
# GET  /agent/info         - Agent metadata
# POST /agent/clear-memory - Clear conversation history
# GET  /agent/health       - Health check

---

📖 Detailed Documentation

Providers

# All providers support the same interface
from aiccel import GeminiProvider, OpenAIProvider, GroqProvider

gemini = GeminiProvider(model="gemini-2.0-flash")
openai = OpenAIProvider(model="gpt-4o-mini")
groq = GroqProvider(model="llama3-70b-8192")

# Generate text
response = gemini.generate("Hello!")

# Chat with history
response = openai.chat([
    {"role": "user", "content": "Hello"},
    {"role": "assistant", "content": "Hi there!"},
    {"role": "user", "content": "How are you?"}
])

# Async variants
response = await gemini.generate_async("Hello!")
response = await openai.chat_async(messages)

Configuration Options

from aiccel import Agent, AgentConfig

config = AgentConfig(
    # Thinking & Reasoning
    thinking_enabled=True,           # Enable chain-of-thought
    thinking_budget=500,             # Max tokens for thinking
    
    # Tool Execution
    strict_tool_usage=False,         # Require tools for all queries
    max_tool_retries=3,              # Retry failed tools
    tool_timeout=30.0,               # Tool execution timeout
    
    # Security
    safety_enabled=True,             # Jailbreak detection
    pii_masking=True,                # Auto-mask PII
    
    # Memory
    memory_type="buffer",            # buffer, window, or summary
    max_memory_turns=50,             # Max conversation turns
    max_memory_tokens=32000,         # Max tokens in memory
)

Memory Types

from aiccel import ConversationMemory

# Buffer Memory (keeps all turns)
memory = ConversationMemory(memory_type="buffer", max_turns=50)

# Window Memory (sliding window)
memory = ConversationMemory(memory_type="window", max_turns=10)

# Summary Memory (summarizes old conversations)
memory = ConversationMemory(
    memory_type="summary",
    max_turns=5,
    llm_provider=provider  # Required for summarization
)

# Use with agent
agent = Agent(provider=provider, memory=memory)

Error Handling

from aiccel.exceptions import (
    AICCLException,       # Base exception
    AgentException,       # Agent-related errors
    ProviderException,    # LLM provider errors
    ToolException,        # Tool execution errors
    ProviderRateLimitError,
    ProviderTimeoutError,
)

try:
    result = agent.run("Query")
except ProviderRateLimitError as e:
    print(f"Rate limited. Retry in {e.retry_after}s")
except ProviderTimeoutError as e:
    print(f"Request timed out after {e.timeout}s")
except ToolException as e:
    print(f"Tool {e.tool_name} failed: {e.message}")
except AgentException as e:
    print(f"Agent error: {e.message}, Context: {e.context}")

---

💡 Real-World Examples

Research Assistant

from aiccel import Agent, GeminiProvider, SearchTool

agent = Agent(
    provider=GeminiProvider(model="gemini-2.0-flash"),
    tools=[SearchTool(api_key="...")],
    name="ResearchAssistant",
    instructions="""You are an expert research assistant.
    - Always cite sources
    - Provide balanced perspectives
    - Summarize key findings clearly"""
)

result = agent.run("What are the pros and cons of nuclear energy?")

Code Assistant

from aiccel import Agent, OpenAIProvider

agent = Agent(
    provider=OpenAIProvider(model="gpt-4o"),
    name="CodeHelper",
    instructions="""You are an expert programmer.
    - Write clean, documented code
    - Explain your implementation
    - Suggest best practices"""
)

# Enable thinking for complex problems
agent.enable_thinking()

result = agent.run("Write a Python function to find the longest palindromic substring")
print(result["thinking"])   # See the reasoning
print(result["response"])   # Get the code

Customer Support Bot

from aiccel import Agent, AgentManager, GeminiProvider

# Specialist agents
billing = Agent(
    provider=GeminiProvider(),
    name="BillingAgent",
    instructions="Handle billing inquiries, refunds, and payment issues."
)

technical = Agent(
    provider=GeminiProvider(),
    name="TechSupport",
    instructions="Troubleshoot technical issues and provide solutions."
)

general = Agent(
    provider=GeminiProvider(),
    name="GeneralSupport",
    instructions="Handle general inquiries and route complex issues."
)

# Router automatically selects the best agent
support_manager = AgentManager(
    llm_provider=GeminiProvider(),
    agents=[billing, technical, general]
)

# Automatic routing
result = support_manager.route("I can't log into my account")  # -> TechSupport
result = support_manager.route("I want a refund")              # -> BillingAgent

---

⚡ Performance Tips

1. Use Connection Pooling (Automatic in v3.2.0+)

# Connections are reused automatically
provider = GeminiProvider()  # Uses connection pooling

2. Use Async for Concurrent Operations

import asyncio

async def process_queries(queries):
    tasks = [agent.run_async(q) for q in queries]
    return await asyncio.gather(*tasks)

3. Enable Caching

from aiccel import Agent, AgentConfig

config = AgentConfig(
    enable_caching=True,
    cache_ttl=3600  # Cache for 1 hour
)

4. Use Lazy Imports

# Fast import (recommended for production)
from aiccel.fast import Agent, GeminiProvider

---

🔧 Environment Variables

Variable	Description	Required
OPENAI_API_KEY	OpenAI API key	For OpenAI
GOOGLE_API_KEY	Google AI API key	For Gemini
GROQ_API_KEY	Groq API key	For Groq
SERPAPI_API_KEY	SerpAPI key	For SearchTool
AICCEL_SECURITY_MODE	FAIL_CLOSED or FAIL_OPEN	Optional

---

🤝 Contributing

We welcome contributions! See our Contributing Guide.

---

📄 License

MIT License - see LICENSE for details.

---

Built with ❤️ for the AI community

⭐ Star us on GitHub • 🐛 Report Issues • 💬 Discussions