pori 1.3

A simple, extensible AI agent framework with tool integration and memory

Pori

Pori is a lightweight, extensible AI agent framework for building intelligent agents with tiered memory, tool-calling, and clean orchestration.

⚡ Quick Start

Installation

Currently, Pori must be installed from source (PyPI publishing is planned — see ROADMAP.md):

git clone https://github.com/aloysathekge/pori.git
cd pori

# Using uv (recommended)
uv venv
.venv\Scripts\activate  # On Windows: .venv\Scripts\activate
                        # On Unix/macOS: source .venv/bin/activate
uv pip install -r requirements.txt

# Or using pip
pip install -r requirements.txt

Configuration

Create config.yaml from config.example.yaml and add your API keys to .env:

cp config.example.yaml config.yaml
# Edit .env with your ANTHROPIC_API_KEY or OPENAI_API_KEY

Basic Usage

Interactive CLI:

python -m pori

Programmatic:

import asyncio
from pori import Orchestrator, AgentSettings, register_all_tools
from pori.llm import ChatAnthropic
from pori.tools.registry import tool_registry
import os
from dotenv import load_dotenv

load_dotenv()

async def main():
    registry = tool_registry()
    register_all_tools(registry)
    
    llm = ChatAnthropic(
        model="claude-sonnet-4-20250514",
        api_key=os.getenv("ANTHROPIC_API_KEY")
    )
    orchestrator = Orchestrator(llm=llm, tools_registry=registry)
    
    result = await orchestrator.execute_task(
        "Calculate the sum of the first 10 Fibonacci numbers",
        agent_settings=AgentSettings(max_steps=10)
    )
    
    if result['success']:
        agent = result.get('agent')
        final_answer = agent.memory.get_final_answer()
        print(f"Answer: {final_answer['final_answer']}")

asyncio.run(main())

Docker

Build and run with Docker:

# Build
docker build -t pori .

# Run (use --env-file to load API keys from .env)
docker run -p 8000:8000 --env-file .env pori

Or with Docker Compose:

# Ensure .env exists with ANTHROPIC_API_KEY
docker compose up --build

Health check: curl http://localhost:8000/v1/health

🧠 Core Features

• Core Memory: Letta-style editable blocks (persona, human, notes) — always in-context
• Custom LLM Wrappers: Direct SDK integration (Anthropic, OpenAI) — no LangChain dependency
• Planning & Reflection: Agent plans tasks and adapts based on results
• Extensible Tools: Simple decorator-based tool registration with Pydantic validation
• Parallel Execution: Orchestrate multiple tasks concurrently
• Comprehensive Logging: Full observability of agent decisions and tool calls

🏗️ Architecture

Pori follows a modular design:

• Orchestrator: Manages task lifecycle, concurrency, and shared memory
• Agent: Core reasoning loop (Plan → Act → Reflect → Evaluate)
• Memory: Conversation history, tool tracking, and Letta-style core memory blocks
• Tool Registry: Validated tool management via Pydantic models
• LLM Wrappers: Lightweight providers (pori/llm/) replacing LangChain

📚 Documentation

• Roadmap — Planned features and contribution areas
• Contributing — How to contribute

📄 License

MIT License.