aiccel 3.1.1

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. Built for developers who need speed (~50ms startup), modularity, and enterprise-grade safety.

---

📦 Installation

AICCEL is modular. Install only what you need.

pip install aiccel             # Core (OpenAI, Gemini, Groq)
pip install aiccel[safety]     # Jailbreak Guard (Transformers)
pip install aiccel[privacy]    # PII Masking (GLiNER)
pip install aiccel[all]        # Full Security & Data Suite

---

🏗️ 1. Building High-Performance Agents

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

from aiccel import Agent, AgentConfig, GeminiProvider

provider = GeminiProvider(api_key="...", model="gemini-2.5-flash")

agent = Agent(
    provider=provider,
    name="Researcher",
    instructions="Find facts and cite sources.",
    config=AgentConfig(
        verbose=True,           # Real-time thought tracing
        safety_enabled=True,    # Active Jailbreak protection
        timeout=30.0            # Fail-safe execution
    )
)

result = agent.run("What is the current state of fusion energy?")

⚙️ Agent Parameters

Parameter	Type	Default	Description
provider	LLMProvider	✅	LLM backend (Gemini, OpenAI, Groq).
name	str	"Agent"	ID for logs and orchestration.
instructions	str	""	The "Soul" - System Prompt & Rules.
tools	List[Tool]	[]	authorized tools for the agent.
config	AgentConfig	None	Operational flags (Timeout, Safety).
memory	Memory	Buffer	Conversation history management.

---

🛡️ 2. Security & Data Privacy Suite

AICCEL provides a multi-layered defense system for AI applications.

🕵️ PII Masking (aiccel.privacy)

Automatically detects and masks sensitive data before it's sent to the LLM.

from aiccel.privacy import mask_text, unmask_text

result = mask_text("Contact John at 555-0123")
# Output: "Contact [PERSON_1] at [PHONE_1]"

# unmask_text(result['masked'], result['mapping']) restores the data.

🛑 Jailbreak Guard (aiccel.safety)

Uses a transformer model to block prompt injection attacks.

• Automatic: Set AgentConfig(safety_enabled=True).
• Manual: if not check_prompt(user_input): abort()

📦 Pandora: Secure Data Analysis

AI-powered ETL agent with process isolation.

from aiccel.pandora import Pandora
# Modes: "local" (fast), "subprocess" (secure), "service" (extreme)
pan = Pandora(provider, execution_mode="subprocess")
df = pan.do(df, "Mask names and calculate revenue growth")

🔐 Secure Vault & Encryption

FIPS-compliant encryption for your secrets.

from aiccel.encryption import SecureVault
vault = SecureVault(master_password="...")
vault.store("STRIPE_KEY", "sk_test_...")

---

🎼 3. Orchestration & Workflows

🤝 Multi-Agent Collaboration

Manage specialist teams using the AgentManager.

from aiccel.manager import AgentManager
manager = AgentManager(llm_provider=p, agents=[researcher, writer])
# Plan-Execute-Synthesize pipeline
response = await manager.collaborate_async("Research and write a report.")

⛓️ Workflow DAGs

Build deterministic agent pipelines.

workflow = (WorkflowBuilder("gen")
    .add_agent("step1", agent1, output_key="data")
    .add_agent("step2", agent2, input_key="data")
    .chain("step1", "step2").build())

---

🧠 4. Advanced Features

• Neural Reranking: Advanced semantic sorting via NeuralReranker.
• Goal Agents: Autonomous agents that pursue complex objectives via GoalAgent.
• Observability: Real-time tracing and structured logging built-in.
• MCP Support: Native Model Context Protocol for cross-platform tools.

---

🔌 API Reference

• Providers: GeminiProvider, OpenAIProvider, GroqProvider.
• Methods: agent.run(), agent.run_async(), agent.stream().

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