Agentix

Build production-ready AI agents with any LLM provider — OpenAI, Claude, Gemini, DeepSeek, or your own endpoint.

Agentix is a modular, pure-Python framework designed to scale from solo developer projects to enterprise-grade deployments, with built-in tools, multi-channel integrations, and advanced reasoning control.

---

🚀 Key Features

• 🧠 LLM-neutral Works with OpenAI, Anthropic (Claude), Google Gemini, DeepSeek, Ollama, vLLM, LiteLLM

• 🔁 True Agent Loop input → LLM → tools → LLM → result

• 🛠 Built-in Tools Filesystem, shell, web search, task management, sub-agents

• 📄 Document Skills (out of the box) PDF, Word, Excel, PowerPoint

• 🌐 Multi-channel Gateway Slack, WhatsApp, Email, Webhooks

• 🧩 Extensible Architecture Skills, plugins, MCP servers, multi-agent orchestration

• ⚙️ Production Ready Logging, sandboxing, hooks, structured output, streaming

---

📦 Installation

pip install agentix-agent-sdk

Optional extras:

pip install agentix-agent-sdk[gateway]   # Slack / WhatsApp / Email
pip install agentix-agent-sdk[all]       # Full install

---

⚡ Quickstart

Simple query

import asyncio
from agentix import AgentixAgentOptions, query, ResultMessage

async def main():
    options = AgentixAgentOptions(
        name="assistant",
        provider="openai",
        model="gpt-4o",
        system_prompt="You are a helpful assistant."
    )

    async for msg in query("What is 2 + 2?", options=options):
        if isinstance(msg, ResultMessage):
            print(msg.result)

asyncio.run(main())

---

Multi-turn agent

import asyncio
from agentix import AgentixAgentOptions, AgentixClient, ResultMessage

async def main():
    options = AgentixAgentOptions(
        name="assistant",
        provider="anthropic",
        model="claude-sonnet-4",
    )

    async with AgentixClient(options) as client:
        async for msg in client.query("Explain Python generators"):
            if isinstance(msg, ResultMessage):
                print(msg.result)

asyncio.run(main())

---

🧠 Core Concepts

Agent Loop

Agentix follows a Claude-style reasoning loop:

User → LLM → Tool Use → LLM → ... → Final Answer

---

🛠 Built-in Capabilities

Tools

• Filesystem: Read, Write, Edit, Search
• Shell: Bash execution
• Web: Search & fetch
• Utility: Task tracking, user input
• Agents: Delegate to sub-agents

Skills

• pdf — extraction, OCR, merge
• docx — Word automation
• pptx — presentation generation
• xlsx — Excel workflows

Some document features require LibreOffice (soffice) in PATH

---

🤖 Multi-Agent Support

from agentix import AgentDefinition

options = AgentixAgentOptions(
    agents={
        "researcher": AgentDefinition(
            description="Search and summarize",
            tools=["WebSearch"]
        )
    }
)

---

🔌 MCP Support

Connect external tool ecosystems via Model Context Protocol:

options = AgentixAgentOptions(
    mcp_servers={
        "filesystem": {
            "type": "stdio",
            "command": "npx",
            "args": ["@modelcontextprotocol/server-filesystem"]
        }
    }
)

---

🌐 Gateway (Optional)

Run agents across:

• Slack
• WhatsApp
• Email
• Webhooks

agentix-gateway

---

⚙️ Configuration

Project-based configuration:

.agentix/
├── settings.json
├── agents/
├── skills/
├── hooks/

---

🔐 Security & Sandbox

• Filesystem isolation
• Controlled tool execution
• Plugin verification

---

📊 Observability

• Structured logging (JSON)
• Tool-level tracing
• Streaming support

---

🔑 Environment Variables

AGENTIX_PROVIDER=openai
AGENTIX_MODEL=gpt-4o
AGENTIX_API_KEY=your-key

---

📚 Documentation

https://docs.agentix.aftws.com/

---

🤝 Contributing

Contributions welcome — please include tests and documentation.

---

📄 License

MIT

---

🔗 Links

• GitHub: https://github.com/aftws-in/agentix-agent-sdk
• Issues: https://github.com/aftws-in/agentix-agent-sdk/issues

---