Back to home@XSakura666

ChronoAgent

Local-first desktop app that schedules AI agent tasks like cron jobs — zero token cost until they run. Write a task, set a time, and it runs automatically with files, web, MCP tools, and multi-model support.

Stars
1
Language
JavaScript
Created
Aug 15, 2026
Updated
Aug 16, 2026

Introduction

⏰ ChronoAgent

Schedule AI agents like cron jobs. Zero token cost until they run.

License: MIT GitHub release Downloads

Write a task now, set a time, and the agent does the work later — automatically.

✨ What is ChronoAgent?

ChronoAgent is a local-first desktop app that turns your backlog into scheduled AI agent runs.

Describe a task in plain text, pick a time, and forget it. Until that moment, ChronoAgent makes zero API calls and costs you zero tokens. When the time comes, it wakes up and runs an AI agent that can read/write files, browse the web, and call any tool you connect via MCP.

scheduler = director · your AI model = brain · your tools = hands

🚀 Features

  • Scheduled execution — tasks run automatically at the time you set
  • 💸 Zero cost before execution — no API calls, no token spend until the timer fires
  • 🤖 Agentic execution — the agent plans, calls tools, and completes multi-step tasks
  • 🧩 Bring your own model — any OpenAI-compatible API (DeepSeek, OpenAI, Ollama, Qwen…)
  • 🔌 Bring your own tools (MCP) — connect any MCP server: filesystem, GitHub, databases…
  • 📦 Local-first & private — tasks and results stay on your machine
  • 🔁 Recurring tasks — daily / weekly / monthly
  • 🪶 Batch execution — merge tasks due at the same time into one call (saves tokens)
  • 🖥️ Runs in the background — tray-resident, auto-start, catches up missed tasks
  • 🧠 Memory catalog — every run auto-archives to a searchable, token-efficient index, so repeat or related tasks can pick up where they left off
  • 🤝 Call other AI & servicesask_model delegates to a second model, http_request calls any HTTP API
  • 🔐 Encrypted API keys — stored encrypted with the OS keychain (DPAPI), never plaintext on disk

🧩 DeepSeek Harness plugin

ChronoAgent is also available as a DeepSeek Harness (DSH) plugin — a host-side scheduled-task agent with the same zero-cost-until-run model, persisted to the session log and recovered across restarts.

See dsh-chrono-agent/ for the plugin source and install guide.

🎬 Demo

🧰 Installation

Download (no Node.js required)

Get the latest build from Releases:

FileType
ChronoAgent Setup.exeInstaller — creates a desktop shortcut
ChronoAgent.exePortable — double-click to run

Windows may show a "SmartScreen" warning (unsigned build). Click More info → Run anyway.

Build from source

git clone https://github.com/XSakura666/ChronoAgent.git
cd ChronoAgent
npm install
npm start

🚦 Quick Start

  1. Add your model — Settings → enter your API base URL + key (a one-click DeepSeek preset is included)
  2. Write a task — e.g. "Write a 500-word product intro and save it to intro.md"
  3. Pick a time — use a quick preset (1 hour, tonight 20:00…) and hit Add
  4. Done — the agent runs at the scheduled time. View results in the task card, or open the workspace folder to see generated files.

🔌 Connect your own tools with MCP

ChronoAgent is an MCP client. Connect any MCP server in Settings → MCP servers:

ToolCommandArgs
Filesystemnpx-y @modelcontextprotocol/server-filesystem D:\data
GitHubnpx-y @modelcontextprotocol/server-github

🏗️ Architecture

┌───────────────────────────────────────────────────┐
│                    ChronoAgent                    │
│                                                   │
│   Scheduler          AI Model           Tools      │
│   (director)         (brain)            (hands)    │
│   · timer            · any OpenAI-      · built-ins │
│   · zero cost        compatible model   · MCP      │
│     until run                            · shell   │
└───────────────────────────────────────────────────┘

🧭 Roadmap

  • Voice input (local transcription — stays zero-cost)
  • MCP template marketplace
  • Code signing (remove SmartScreen warning)
  • Linux / macOS builds

🤝 Contributing

Pull requests and ideas are welcome.

📄 License

MIT