Connecting an AI Tool

TandemLoop doesn’t include its own AI. The app provides the ground rules for working together—the AI itself is an external, agentic tool of your choice that you start separately and give access to your data folder.

TandemLoop Doesn’t Supply an AI—You Bring Your Own

This is a deliberate design decision: TandemLoop contains no built-in AI service, doesn’t contact third-party servers, and doesn’t send your data anywhere. Instead, the data folder—your folder of Markdown files—is the shared workspace. Any AI tool that can read and write this folder can work with you in tandem.

For working with AI, you therefore always need an external tool. Without such a tool, TandemLoop works fully—just without AI.

Which Tools You Can Use

Any agentic AI tool that can access a local folder and read and write files is suitable. These include, for example:

• Claude Cowork and Claude Code (Anthropic)
• Codex (OpenAI)
• Gemini CLI and Antigravity (Google)

The list isn’t exhaustive—what matters isn’t the name of the tool but that it can use your data folder as its working directory and create files there. TandemLoop doesn’t favor any particular tool and doesn’t hard-wire any.

Quickest Way: The Agent Tutorial (Mac)

On the Mac, the Agent Tutorial walks you through working with your agent in four exercises. It appears after initial setup if you answer “Do you work with AI agents?” with Yes and choose Start Tutorial—and it’s available anytime from Help › Agent Tutorial or Settings › AI Agents › Start Agent Tutorial.

For each exercise, the tutorial shows a ready-made prompt. You copy it into your agent, and the tutorial detects live when your agent has created the files in the data folder.

1. A task in the Inbox. Your agent reads AGENTS.md, looks up the TandemLoop website, and adds a task with the link to your Inbox.
2. Plan a project. For this exercise, TandemLoop creates the TandemLoop Onboarding area. Your agent plans a project there with milestones and tasks covering views, specs, handbacks, and the Weekly Review.
3. Build a view. Your agent creates the Next Steps view—all projects, with only the next task from each—and marks the matching task in the onboarding project as completed.
4. Get work back. A handback is waiting in the demo area: your agent has drafted an onboarding plan. In Back from Agent, you give feedback with Request Changes, let your agent know, and it hands the revised plan back to you. Then you accept it.

If proposals are turned on, your agent submits the project and the view as proposals; they appear only after you approve them.

At the end, you’ll find the worker skill to keep your agent working, ideas for working together, and a prompt your agent uses to link the data folder into a project. If you select the checkbox before Finish Tutorial, TandemLoop moves the TandemLoop Onboarding area and all its contents to the Trash; without the checkbox, it stays. The Inbox task and the view stay in any case.

The introduction and the final page appear in the center of the window, the exercises in a panel on the right edge of the window—you can keep using the app. Click – to minimize the panel and ✕ to close it; the next time you open it, it picks up where you left off.

How to Connect a Tool

At its core, connecting a tool means pointing it to your data folder and referring it to the included instructions.

1. Find the path to the data folder. In TandemLoop, open Settings and look under Data Folder to see where it is.
2. Point the tool at the folder. Start your AI tool and give it this folder as its working directory or access scope.
3. Refer it to AGENTS.md. The root of the data folder contains the file AGENTS.md. It explains the rules to the tool: permitted actions, the folder structure, and—if turned on—the proposal format. Ask your tool to read it first.
4. Give it a task. Describe what the AI should do—for example, “clean up my Inbox” or “structure project X into milestones.”

If you’ve turned on proposals (Settings › AI Agents, off by default), a well-instructed tool then saves its proposals in the _proposals/ folder. TandemLoop detects them automatically and shows them to you for a decision—How Proposals Are Created describes how this works in detail. Without proposals, the tool creates and changes items directly; it deletes only when you explicitly ask it to.

A concrete task helps you get started. Copy this prompt, replace <DATA-FOLDER-PATH> (from Settings), and give it to your tool:

My TandemLoop data folder is at <DATA-FOLDER-PATH>. First read the file
AGENTS.md in the root of the data folder and follow the rules there. Then go
through my Inbox and suggest how the items could be sorted into suitable
projects.

Depending on the setting, the tool saves its suggestion as a proposal or describes it to you in the conversation and carries it out after you agree.

Trigger It Once or Keep It Running

You determine how close the collaboration is through the way you run your tool:

• On request. You start the tool, give it a task, review the result—and close it again. Ideal for targeted cleanup or planning runs.
• Continuously in the background. You keep a tool running so that it regularly goes through the data folder, works on tasks assigned to it, and responds to your feedback. This way, the AI becomes a partner that runs alongside you.

In both cases, you stay in control: The agent may delete, or complete or drop your tasks, only at your request, and with proposals it changes nothing structural until you approve it. Working Together Autonomously & Continuously shows how to set up continuous operation with the included worker skill—on request or in a loop.