AgentOS (fikriaf/agentos)

Autonomous AI agent framework from https://github.com/fikriaf/agentos

Installation

git clone https://github.com/fikriaf/agentos.git /opt/agentos
cd /opt/agentos
pip install -e .

Requirements

Python 3.10+
OpenAI API key (or compatible)
Skills in ~/.hermes/skills/

Configuration

Set environment variable:

export OPENAI_API_KEY="sk-..."

Or use config file at ~/.agentos/config.yaml:

llm:
  provider: "openai"
  model: "gpt-4"
  api_key: "${OPENAI_API_KEY}"

Commands

# Run task (auto mode)
agentos run "Build a REST API for todo list"

# Interactive wizard mode
agentos wizard

# Session management
agentos sessions          # List sessions
agentos resume --session session_id
agentos delete --session session_id

# Info
agentos info

Quick mode with auto-confirm:

agentos run "Write tests" --auto-confirm

Architecture

User Input → Context Engineering → ROMA Planner → HTAA Executor → MOSAIC Safety → Memory → Reflection

Components

ROMA Planner - Recursive task decomposition into parallel subtasks
HTAA Executor - Tool grouping for reduced action space
MOSAIC Safety - Plan → Check → Act/Refuse safety pipeline
Memory - File-based state externalization (InfiAgent)
Skills - 93 built-in skills loaded on demand

Key Files

src/agentos/cli.py - Main CLI entry point
src/agentos/agents/planner.py - ROMA planner implementation
src/agentos/agents/executor.py - HTAA executor with tool grouping
src/agentos/skills/__init__.py - SkillsManager loader
SPEC.md - Full technical specification

Skills Integration

Copy existing skills to AgentOS:

cp -r ~/.hermes/skills/* /opt/agentos/src/agentos/skills/
cd /opt/agentos && pip install -e .

AgentOS loads skills via skill_view(name) for context.

Known Issues

Issue	Solution
Safety blocks startup	Use `--auto-confirm` flag
Python tool fails	Use `python3` not `python` in commands

Troubleshooting

# Debug mode
agentos run "task" --verbose

# Check version
agentos --version

# Show config
agentos info

agentos