easylocai 0.1.0

Local AI CLI using local llm gpt-oss:20b

easylocai

Fully autonomous agentic workflows running locally—no APIs, no data leaks

Overview

Easylocai is an On-Device Autonomous Agent designed for secure, offline task execution. Unlike cloud-dependent assistants, it leverages the gpt-oss:20b model to perform complex reasoning and actions entirely on your local machine.

By implementing a sophisticated Plan-Execute-Replan orchestration, Easylocai can decompose ambiguous goals into actionable steps, execute them using Model Context Protocol (MCP) tools, and autonomously refine its strategy based on real-time feedback.

Features

• Privacy-First Autonomy: 100% local execution using gpt-oss:20b via Ollama. Your code and data never leave your machine.

• Agentic Orchestration: A robust multi-agent loop (Plan → Execute → Replan) that ensures high success rates for long-horizon tasks.

• MCP Tool Integration: Seamlessly connects with Model Context Protocol (MCP) servers to interact with your local file system, terminal, and APIs.

Requirements

To ensure stable performance of the autonomous agent, your system must meet the following criteria:

System Requirements

• Minimum 16GB RAM (32GB or more recommended for optimal performance)
• Sufficient disk space for model storage and operation

OS

• OS: macOS (Strictly supported)

Software Requirements

• Runtime: Python 3.12. It is recommended to use pyenv.
• LLM Engine: Ollama must be installed and running.
  • Model: gpt-oss:20b (Make sure to run ollama pull gpt-oss:20b before starting).

Install & Execution

Installation

bash install.sh

Configuration

MCP server configuration

• file_name: ~/.config/easylocai/config.json
• example

  {
    "mcpServers": {
      "filesystem": {
        "command": "npx",
        "args": [
          "-y",
          "@modelcontextprotocol/server-filesystem",
          "."
        ]
      },
      "kubernetes": {
        "command": "python",
        "args": [
          "-m",
          "kubectl_mcp_tool.mcp_server"
        ],
        "cwd": "~/Programming/kubectl-mcp-server",
        "env": {
          "KUBECONFIG": "~/.kube/config",
          "KUBECTL_MCP_LOG_LEVEL": "ERROR",
          "PYTHONUNBUFFERED": "1"
        }
      },
      "notion_api": {
        "command": "docker",
        "args": [
          "run",
          "--rm",
          "-i",
          "-e", "NOTION_TOKEN",
          "mcp/notion"
        ],
        "env": {
          "NOTION_TOKEN": "<token>"
        }
      }
    }
  }
  

Initialization

easylocai init

If you want to force re-initialization, use --force flag:

easylocai init --force

Run

easylocai

Run beta workflow variant

easylocai --flag=beta

Development

Run without installing

python -m easylocai.run

# Beta variant
python -m easylocai.run --flag=beta

• refer to ARCHITECTURE.md