Introducing AI Agent Platform: Schedule and Chain Any AI Agent on Windows
Like Task Scheduler grew a brain, got a Windows 11 makeover, and started talking to GPT.
If you’ve spent any time building AI agents, you’ve probably noticed a familiar pattern: a Python script here, a Streamlit app there, a cron job cobbled together somewhere else, and a pile of API keys scattered across .env files you’ve half-forgotten about. It works — until it doesn’t. Updating a prompt means opening a terminal. Changing a schedule means editing a crontab. Checking whether an agent actually ran means digging through logs.
I wanted something better. Something that felt like a real Windows application — not a browser tab, not a terminal window, not a dependency manager waiting to break. So I built it.
AI Agent Platform is a standalone Windows 11 desktop application that lets you create, schedule, and monitor AI agents against any major LLM provider — with nothing to install beyond the app itself.
Why It Matters
The gap between “I can use AI” and “AI is working for me automatically” has always been the infrastructure in the middle. Scheduling, persistence, credential management, output capture, monitoring — none of that is hard in isolation, but assembling it from scratch for every new agent idea is friction that stops most people from building anything durable.
AI Agent Platform removes that friction entirely. It treats AI agents as first-class scheduled tasks — the same mental model as a Windows Task Scheduler job or a cron entry, but purpose-built for LLMs with a UI that actually reflects what you’re configuring. Non-developers can use it. Developers don’t have to waste time on plumbing.
It’s also genuinely portable. Because all user data lives in Documents\AIAgentPlatform\, OneDrive — or any sync service — roams your agents, settings, and keys across every machine you own automatically. There’s no “export your config” step. You just install the app on a second machine and everything is already there.
Why It’s Cool
It runs on Snapdragon. The installer is a universal binary — drop it on an ARM64 Copilot+ PC and it just works, natively, at full speed. That matters because the machines most likely to be always-on, always-connected Windows devices right now are ARM-based.
The whole thing is built without a backend. No server, no database, no API of its own — just an Electron app talking directly to whichever LLM you told it to use. The app itself has no idea what your prompts say or what your agents produce. It’s a scheduler and a UI. The intelligence is entirely yours, pointed wherever you want.
And the agent chaining feature quietly unlocks something significant. The moment you can say “take the output of this agent and feed it into that agent,” you’re not scheduling prompts anymore — you’re composing workflows. A news digest that feeds a social post drafter. A market snapshot that feeds a portfolio commentary agent. A code review agent that feeds a ticket-writing agent. All of it running on a timer, producing finished output, without you in the loop. That’s genuinely useful automation, and it took about a dozen lines of code to implement because the platform’s architecture made it a natural extension rather than a retrofit.
It’s a real Windows app, built to feel like one, that makes AI do your work while you’re not watching. That’s the whole idea — and it works.
What It Does
At its core, AI Agent Platform is a scheduling engine for two kinds of agents:
Prompt Agents connect to an LLM — any LLM — with a system prompt and a user prompt, run on a schedule you define, and capture the response directly in the app. Think daily summaries, automated research briefings, content drafts, data analysis, or anything else you’d normally send to a chatbot by hand, now running automatically on a timer.
Script Agents run any existing script or executable — Python, PowerShell, Node.js, a compiled .exe, whatever you already have — on the same cron-based scheduler. If you’ve already built agents in another environment, you can import them here without rewriting a line of code.
Both types display live status, last-run timestamps, and captured output directly on their cards in the main window. No log files. No terminals. Just results.
You define an agent in one of two ways. A Prompt Agent is a system prompt plus a user prompt pointed at a model — the kind of thing you’d normally type into a chat window, now running on a schedule and saving its output automatically. A Script Agent runs any existing script or executable you already have — Python, PowerShell, Node.js, anything — on the same scheduler. Both types live on cards in the main window with live status indicators, run history, and captured output.
From there, the platform does a lot of lifting:
Schedules run on cron, with a visual builder so you don’t need to memorize the syntax
Output history persists every result, not just the last one, so nothing gets overwritten and lost
Agent chaining lets you pipe one agent’s output directly into another’s prompt — multi-step AI pipelines with zero code
Native Windows notifications tell you when something finishes or fails without you having to watch the app
Community packs give you a gallery of ready-to-install agent collections — cybersecurity briefings, market snapshots, daily learning flashcards, productivity plans — imported with a single click
Test mode lets you run a configuration before you save it, so you’re not committing to something you haven’t validated
Open as HTML renders any markdown output as a clean, styled page in your browser — because a 500-word AI briefing deserves better than a monospace text box
Provider Agnostic by Design
One of the core design goals was to avoid locking the platform to any single AI provider. The app ships with built-in support for five providers — and switching between them is a matter of clicking a chip and pasting an API key:
Each agent independently selects its provider and model, so you can run a Claude agent for summarization, a Grok agent for research, and an Ollama agent for a local task — all on the same scheduler, at the same time.
A Windows App That Actually Looks Like a Windows App
The UI is built on the Windows 11 Fluent Design System — Mica background material, Acrylic blur layers, Fluent motion tokens, and Segoe UI Variable throughout. It feels native because it’s designed to be.
The sidebar handles everything about creating or editing an agent. The Prompt tab exposes the full LLM configuration — system prompt, user prompt, provider, model, temperature, and schedule:
Configuring a prompt-based agent with provider and model selection.
The Script tab is for existing automation. Point it at any script, set a command interpreter and a timeout, pick a schedule:
Configuring a script-based agent — Python, PowerShell, Node.js, or any executable.
Settings That Stay Out of Your Way
The Settings dialog handles provider configuration, system tray behavior, Windows startup, and notification preferences — all in one place:
Per-provider API key management, tray, startup, and notification toggles — all in one dialog.
A few things worth noting about how credentials are handled:
Keys are stored locally in
Documents\AIAgentPlatform\settings.json— a plain file on your machine, under your controlKeys are never transmitted anywhere except directly to the provider’s API endpoint you configured
The masked hint (e.g.
xai-SanS...lWxH) lets you verify which key is stored without exposing it
The Minimize to System Tray toggle keeps agents running even when you close the window — the scheduler stays alive in the background, and the tray icon gives you one-click access to open the app or stop the scheduler. The Run at Windows Startup toggle writes directly to the Windows registry so your agents are ready to go from the moment you sign in.
How the Scheduler Works
Scheduling uses standard cron expressions. The app ships with a dropdown of the most common presets — every 5 minutes, every hour, daily at 8 AM, weekdays at 9 AM — plus a “Custom cron…” option backed by the new visual builder (more on that below). Behind the scenes, node-cron handles the scheduling with overlap prevention built in: if an agent is still running when its next tick fires, the new run is skipped rather than stacked.
When the scheduler is running, the pill in the header shows the count of active agents and updates in real time as agents complete, fail, or are added and removed.
No Cloud, No Runtime, No Drama
The app is a single installable .exe — built with Electron and packaged with electron-builder into a standard NSIS installer that supports both x64 and ARM64 (including Snapdragon-based Windows PCs). The installer auto-detects your architecture and installs the right version. After installation:
No Node.js required on the target machine — the runtime is bundled
No Python required — no virtual environments, no
pip install, no version conflictsNo internet connection required to run the app itself — only the actual LLM API calls go out, to the provider you chose
All data lives in
Documents\AIAgentPlatform\— survives app updates and reinstalls, and is easy to back up
Documents\AIAgentPlatform\
├── agent_registry.json ← agent definitions and latest run state
├── settings.json ← provider config and app preferences
└── history\
└── {agentId}.json ← per-agent run history (up to 50 entries each)
There’s a side effect of that location worth calling out: because the data folder sits inside Documents\, it is automatically synced by OneDrive — and any other folder-sync service you use. Install AI Agent Platform on a second machine, sign in to the same OneDrive account, and your agents, settings, and provider keys are already there before you open the app for the first time. No export, no copy, no reconfiguration.
Share Your Agents
One of the most requested things in any automation tool is the ability to share your work. AI Agent Platform makes that simple: every agent you create can be exported to a plain JSON file and imported by anyone else running the app.
Hit ⬇ Export in the Active Agents header and a native Save dialog appears. The file contains everything needed to recreate your agents — names, prompts, providers, models, schedules — with runtime state and API keys intentionally excluded.
On the receiving end, ⬆ Import opens a file picker, reads the JSON, and adds each new agent to the local registry. If an agent with the same name already exists it is skipped rather than overwritten, and the import toast tells you exactly what happened:
Imported 3 agents. Skipped 1 duplicate (WeatherBot).
What’s New in v1.1.0
V1.1.0 ships nine new features based on feedback from early users. Here’s what changed.
Output History
The app now persists every run result — not just the most recent one. Each agent stores up to 50 entries in its own history file under Documents\AIAgentPlatform\history\. Click the History button on any agent card to open a scrollable dialog showing every past run with its timestamp, status, and full output. Each entry has its own 📋 Copy and 🌐 Open as HTML buttons.
Windows Toast Notifications
Agents now fire native Windows toast notifications when they complete or fail — the same system notifications you get from any other Windows app, appearing in the notification center and as a pop-up in the corner of the screen. There’s a toggle in Settings to turn them off if you prefer a quieter experience.
Agent Chaining
This one changes how you think about what agents can do. Every agent now has an optional Chain To field. Set it to another agent and, after a successful run, the first agent’s output is automatically injected as the userPrompt of the second agent — which then triggers immediately.
The practical upshot: multi-step pipelines with no code. A News Summarizer can feed directly into a Tweet Drafter. A Market Snapshot can chain into an Email Formatter. The chain badge on the card shows you at a glance which agent fires next.
Import from Clipboard
Creating an agent from a prompt you’ve been refining in a chat session used to mean copying it, switching apps, finding the field, and pasting. Now there’s a 📋 Paste button right next to the User Prompt field. One click and your clipboard contents land in the form.
Test Before Saving
The ▶ Test button in the sidebar runs your current form configuration once without saving anything. The result — or the error — appears in a panel below the buttons immediately. Iterate on your prompt, test again, save when it’s right. No more creating an agent, running it, checking the output, deleting it, and starting over.
Copy Output and Open as HTML
Two new buttons on every agent card:
📋 copies the latest output to the clipboard in one click
🌐 renders the output as a styled HTML page and opens it in your default browser
The HTML renderer converts markdown headings, lists, code blocks, bold, italic, and links into clean, readable HTML with a light theme — no external dependencies, no Markdown library to ship. It’s useful any time an agent produces a long formatted response that’s easier to read in a browser than in the app’s output panel.
Enhanced Cron Builder
“Custom cron…” used to drop you straight into a text box. Now it opens a two-tab panel:
Builder — choose a frequency pattern (every N minutes, every N hours, daily, weekly, or monthly), pick a time and day from dropdowns, and watch the cron expression and a plain-English description update live. No cron syntax knowledge required.
Expression — the original raw text input, for users who already know what 0 9 * * 1-5 means and just want to type it.
Both tabs stay in sync. Switch from Builder to Expression and the generated expression is already there.
Community Agent Packs
The 🌐 Packs button in the Active Agents header opens a gallery that fetches ready-to-import agent packs from the project’s GitHub repository. Each pack is a curated collection of agents around a common theme. Click Install and the entire pack is imported in one step.
Six packs ship with v1.1.0:
The packs index is a plain JSON file in the repo — adding a new pack is a pull request.
Get It
The project is open source under the MIT license.
GitHub: https://github.com/rod-trent/AgentPlatform
To run from source:
git clone https://github.com/rod-trent/AgentPlatform.git
cd AgentPlatform
npm install
npm startTo build the installer:
npm run buildThe installer lands in dist\AI Agent Platform Setup 1.1.0.exe.
What’s Next
A few things already on the list for v1.2.0:
Conditional logic — only trigger a chained agent if the upstream output matches a pattern or keyword
Agent groups — tag agents into logical groups and start/stop an entire group at once
Run on demand from the tray — right-click any agent directly from the system tray icon and run it without opening the main window
Output diffing — highlight what changed between the current and previous run result
Webhook triggers — fire an agent in response to an incoming HTTP request instead of (or in addition to) a schedule
If you build something with it, run into a bug, or have a feature idea, open an issue on GitHub. Pull requests welcome.
AI Agent Platform is open source software released under the MIT License. Copyright © 2025 Rod Trent.