aiccel 3.1.0

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

AICCEL Framework 3.1: The Production-Grade Agentic Library

AICCEL (AI-Accelerated Agentic Library) is a security-first, high-performance framework for building orchestrated AI systems. Designed for developers who need speed (~50ms startup), modularity, and enterprise-grade safety.

---

📦 Installation

AICCEL is modular. Install only what you need to keep your deployment lightweight.

# 🚀 Core Framework (Lightweight)
pip install aiccel

# 🛡️ Production Security (Jailbreak Detection & Transformers)
pip install aiccel[safety]

# 🕵️‍♀️ Privacy Suite (PII Masking with GLiNER)
pip install aiccel[privacy]

# 📦 The Full Suite (Recommended for Dev)
pip install aiccel[all]

---

🏗️ Building Agents

The Agent class handles LLM interaction, tool execution, and session memory.

Quick Start

from aiccel import Agent, GeminiProvider, AgentConfig

# 1. Setup
provider = GeminiProvider(api_key="...", model="gemini-2.5-flash")
config = AgentConfig(safety_enabled=True, verbose=True)

# 2. Initialize
agent = Agent(
    provider=provider, 
    config=config,
    instructions="You are a helpful analyst."
)

# 3. Run (Sync or Async)
result = agent.run("Summarize the impact of AI on medicine.")
print(result['response'])

Advanced Configuration (AgentConfig)

Flag	Default	Description
verbose	False	Enable detailed thinking & tool logs.
timeout	60.0	Max seconds per generation.
safety_enabled	False	Enable transformer-based Jailbreak guard.
lightweight	False	Disable heavy model loading for serverless/edge.

---

🎼 Orchestration: The Agent Manager

Manage multiple specialized agents using the Plan-Execute-Synthesize workflow.

from aiccel.manager import AgentManager

# 1. Initialize specialized agents
researcher = Agent(name="Researcher", description="Finds facts.")
writer = Agent(name="Writer", description="Writes articles.")

# 2. Orchestrate
manager = AgentManager(llm_provider=provider, agents=[researcher, writer])
response = await manager.collaborate_async("Research fusion energy and write a summary.")

---

🛡️ Security & Privacy Suite

1. Jailbreak Guard

Automatically blocks malicious prompts designed to bypass instructions.

# Enabled via config.safety_enabled = True
# Uses 'traromal/AIccel_Jailbreak' model

2. PII Masking

Protects sensitive data (Email, Phone, Names) from leaving your server.

from aiccel.privacy import mask_text, unmask_text

result = mask_text("Contact John at [e-mail]")
# Output: "Contact [PERSON_1] at [EMAIL_1]"

3. Pandora: Isolated Execution

AICCEL's data analysis engine (Pandora) supports three isolation modes:

• local: Restricted builtins (Fast).
• subprocess: Fresh OS process (Medium Isolation).
• service: Remote microservice (Maximum Security).

from aiccel.pandora import Pandora
pan = Pandora(provider, execution_mode="subprocess")
df = pan.do(df, "Calculate growth rate and mask customer names")

---

🔀 Workflow DAGs

Build deterministic pipelines using a Directed Acyclic Graph.

from aiccel import WorkflowBuilder, WorkflowExecutor

workflow = (
    WorkflowBuilder("data_pipeline")
    .add_agent("fetch", agent1, output_key="raw_data")
    .add_agent("process", agent2, input_key="raw_data")
    .chain("fetch", "process")
    .build()
)

result = await WorkflowExecutor().run(workflow, inputs)

---

🔌 API Reference & Extensibility

Supported Providers

• GeminiProvider (Google)
• OpenAIProvider (OpenAI)
• GroqProvider (Llama 3.1, Mixtral)

Custom Tools

Subclass BaseTool to extend your agents' capabilities with JSON-schema validation.

---

Built for speed, built for safety. Built by the AICCEL Team.