Openhanako Personal AI Agent

OpenHanako Personal AI Agent

Skill by ara.so — Daily 2026 Skills collection.

OpenHanako is a desktop AI agent platform built on Electron that gives each agent persistent memory, a distinct personality, and the ability to autonomously operate your computer — read/write files, run terminal commands, browse the web, execute JavaScript, and manage schedules. Multiple agents can collaborate via channel group chats or task delegation.

Installation

Download & Run

# macOS Apple Silicon — download from releases page
# https://github.com/liliMozi/openhanako/releases
# Mount the .dmg and drag to Applications

# First launch — bypass Gatekeeper (one-time):
# Right-click app → Open → Open

# Windows — run the .exe installer from releases
# SmartScreen warning: click "More info" → "Run anyway"

Build from Source

git clone https://github.com/liliMozi/openhanako.git
cd openhanako
npm install

# Development mode
npm run dev

# Build for production
npm run build

# Run tests
npm test

First-Run Onboarding

On first launch, the wizard asks for:

1. Language — UI language preference
2. Your name — used by agents when addressing you
3. Model provider — any OpenAI-compatible endpoint
4. Three models:

• chat model — main conversation (e.g. gpt-4o, deepseek-chat)
• utility model — lightweight tasks, summarization (e.g. gpt-4o-mini)
• utility large model — memory compilation, deep analysis (e.g. gpt-4o)

Provider Configuration Examples

// OpenAI
{
  "baseURL": "https://api.openai.com/v1",
  "apiKey": "process.env.OPENAI_API_KEY"
}

// DeepSeek
{
  "baseURL": "https://api.deepseek.com/v1",
  "apiKey": "process.env.DEEPSEEK_API_KEY"
}

// Local Ollama
{
  "baseURL": "http://localhost:11434/v1",
  "apiKey": "ollama"
}

// Qwen (Alibaba Cloud)
{
  "baseURL": "https://dashscope.aliyuncs.com/compatible-mode/v1",
  "apiKey": "process.env.DASHSCOPE_API_KEY"
}

Project Architecture

openhanako/
├── core/           # Engine orchestration + Managers (Agent, Session, Model, Preferences, Skill)
├── lib/            # Core libraries
│   ├── memory/     # Custom memory system (recency decay)
│   ├── tools/      # Built-in tools (files, terminal, browser, screenshot, canvas)
│   ├── sandbox/    # PathGuard + OS-level isolation (Seatbelt/Bubblewrap)
│   └── bridge/     # Multi-platform adapters (Telegram, Feishu, QQ)
├── server/         # Fastify 5 HTTP + WebSocket server
├── hub/            # Scheduler, ChannelRouter, EventBus
├── desktop/        # Electron 38 main process + React 19 frontend
├── tests/          # Vitest test suite
└── skills2set/     # Built-in skill definitions

Key Managers (via unified engine facade)

Agent Configuration

Each agent is a self-contained folder you can back up:

~/.openhanako/agents/<agent-id>/
├── personality.md      # Personality template (free-form prose or structured)
├── memory/
│   ├── working.db      # Recent events (SQLite WAL)
│   └── compiled.md     # Long-term compiled memory
├── desk/               # Agent's file workspace
│   └── notes/          # Jian notes
└── skills/             # Agent-local installed skills

Personality Template Example

# Hanako

You are Hanako, a calm and thoughtful assistant who prefers directness over verbosity.
You remember past conversations and refer to them naturally.
You ask clarifying questions before starting large tasks.
When writing code, you always add brief inline comments.

## Tone
- Warm but professional
- Uses occasional dry humor
- Never uses hollow affirmations ("Great question!")

## Constraints
- Always confirm before deleting files
- Summarize long terminal output rather than dumping it raw

Skills System

Skills extend agent capabilities. They live in skills2set/ (built-in) or are installed per-agent.

Install a Skill from GitHub

// Via the Skills UI in the app, or programmatically:
const { skillManager } = engine;

await skillManager.installFromGitHub({
  repo: 'some-user/hanako-skill-weather',
  agentId: 'agent-abc123',
  safetyReview: true   // strict review enabled by default
});

Skill Definition Format (SKILL.md → skills2set)

---
name: web-scraper
version: 1.0.0
description: Scrape structured data from web pages
tools:
  - browser
  - javascript
permissions:
  - network
---

## Instructions for Agent

When asked to scrape a page:
1. Use the `browser` tool to navigate to the URL
2. Use `executeJavaScript` to extract structured data
3. Save results to the desk as JSON

Writing a Custom Skill (JavaScript)

// skills/my-skill/index.js
export default {
  name: 'my-skill',
  version: '1.0.0',
  description: 'Does something useful',

  // Tools this skill adds to the agent
  tools: [
    {
      name: 'fetch_weather',
      description: 'Fetch current weather for a city',
      parameters: {
        type: 'object',
        properties: {
          city: { type: 'string', description: 'City name' }
        },
        required: ['city']
      },
      async execute({ city }) {
        const res = await fetch(
          `https://wttr.in/${encodeURIComponent(city)}?format=j1`
        );
        const data = await res.json();
        return {
          temp_c: data.current_condition[0].temp_C,
          description: data.current_condition[0].weatherDesc[0].value
        };
      }
    }
  ]
};

Memory System

OpenHanako uses a recency-decay memory model: recent events stay sharp, older ones fade.

// Accessing memory programmatically (core/lib/memory)
import { MemoryManager } from './lib/memory/index.js';

const memory = new MemoryManager({ agentId: 'agent-abc123' });

// Store a memory event
await memory.store({
  type: 'conversation',
  content: 'User prefers dark mode and terse responses',
  importance: 0.8   // 0.0–1.0; higher = decays slower
});

// Retrieve relevant memories
const relevant = await memory.query({
  query: 'user preferences',
  limit: 10,
  minRelevance: 0.5
});

// Trigger manual compilation (normally runs automatically)
await memory.compile();

Memory Tiers

Built-in Tools

Tools available to agents out of the box:

// File operations
{ tool: 'read_file',   args: { path: '/Users/me/notes.txt' } }
{ tool: 'write_file',  args: { path: '/Users/me/out.txt', content: '...' } }

// Terminal
{ tool: 'run_command', args: { command: 'ls -la', cwd: '/Users/me' } }

// Browser & web
{ tool: 'browse',      args: { url: 'https://example.com' } }
{ tool: 'web_search',  args: { query: 'OpenHanako latest release' } }

// Screen
{ tool: 'screenshot',  args: {} }

// Canvas
{ tool: 'draw',        args: { instructions: '...' } }

// Code execution
{ tool: 'execute_js',  args: { code: 'return 2 + 2' } }

Sandbox Access Tiers (PathGuard)

Tier 0 — Denied:     System paths (/System, /usr, registry hives)
Tier 1 — Read-only:  Home directory files outside agent desk
Tier 2 — Read-write: Agent desk folder only
Tier 3 — Full:       Explicitly granted paths (user confirms)

OS-level sandbox: macOS Seatbelt / Linux Bubblewrap wraps the skill process.

Multi-Agent Setup

// core/AgentManager usage example
import { createEngine } from './core/engine.js';