How to use LoopEngine

LoopEngine runs AI agents that do routine work for you, such as answering customer emails, triaging alerts or checking expense claims. Each agent works in the apps you connect, and anything risky, like a refund or a message to a customer, waits for a person to approve it first.

Sign up, then create a workspace from your dashboard. A workspace is your own LoopEngine, at its own address (your-name.loopengine.app), where your agents, apps and settings live. Once it's ready, open it from the dashboard; you're signed in to it automatically.

The one you choose. Open Models in the workspace's sidebar and save a key for OpenAI, Anthropic, Gemini or another provider. Each agent then picks its provider and model on its Overview tab, under Model. For OpenAI and Anthropic models you can also set how much the model thinks before answering (Reasoning effort): more is slower but handles harder cases better.

The sidebar has four ways:

• Find your first agent asks you a few questions about your work and suggests one.
• Start from an example sets up a ready-made workflow for a common job, with its apps and permissions. See the workflow examples.
• Build a workflow, on an agent's Workflows tab, turns a job you describe in a sentence or two into a workflow for that agent. See Workflows.
• + New next to Agents starts a blank agent you set up yourself.

Open the agent and use its Chat tab: ask it something the way a colleague or customer would. You see each step it takes, and anything that needs your approval stops there for you to decide.

Its instructions are on the Overview tab: plain text describing its job, like a brief for a new colleague. Edit them and save; the next conversation uses the new version. The other tabs set what it can use (Apps), what it may do on its own (Permissions), where it asks you (Messaging) and when it runs (Schedules).

Add a schedule on its Schedules tab, such as every 10 minutes or every Monday at 9. New schedules start paused, so nothing runs until you turn them on.

Settings the agent's apps or abilities still need, usually an API key or token for a service it works with. Open Environment, fill them in, and the count goes away. Values are stored in your workspace and never shown again.

Open Apps in the sidebar, choose the app (Stripe, GitHub, Notion, Linear and many more), and click Connect. Most apps show Sign in with the app: you approve LoopEngine on the app's site and come back connected. Apps that take a key, like GitHub, ask you to paste it instead. Either way it's stored in your workspace. See Apps.

Connecting an app gives no agent anything. Open the agent's Apps tab, add the app, and tick the tools this agent should have. Choosing only the read-only tools is a safe place to start; you can change an agent's tools any time without affecting other agents.

Ready-made sets of tools for things that aren't apps with their own connection, such as searching the web or your documents, and your own company's systems. Install them from Abilities in the sidebar, then add them to an agent from its Apps tab. Private abilities can come from your company's GitHub: add the organization and a token that can read its repositories, and they appear on the Abilities page.

Every tool has a permission, set on the agent's Permissions tab (or as you add it on the Apps tab):

• Allow: runs on its own. For safe, routine actions such as looking up an order.
• Ask me: waits until a person approves, edits or denies it. For refunds, payments and anything sent to a customer.
• Never: blocked, whatever the agent decides.

Wherever you set on the agent's Messaging tab: Slack, Lark, email or a webhook. The first messaging app you save becomes the primary one. Each request says what the agent wants to do and why, with Approve and Deny buttons.

That piece of work pauses; nothing is done until someone decides. Approve it and the agent carries on from where it stopped; deny it and the agent is told it can't, and works out what to do instead. You can also decide from the agent's page in the console.

Create an API key on your workspace's page in the dashboard, under API keys. Then send the agent a message:

POST https://your-name.loopengine.app/agents/<agent-name>/messages
Authorization: Bearer lek_...
Content-Type: application/json

{ "message": "Where is order EM-1001?" }

The reply is in the response's text. If the agent stopped to ask for an approval, the response is 202 with pending: true instead.

Treat a key like a password: keep it on your server, never in a web page or app people download. Create one key per system that calls your workspace, so you can revoke one from the dashboard without affecting the others.

Open Settings in the sidebar. When a new version is out, it offers to install it; that takes a minute or two, and your agents, apps and settings stay as they are.

From its page in the dashboard, under Delete workspace. This removes its agents, connected apps, settings and history, and can't be undone.

If something isn't working or isn't covered here, open an issue on GitHub. Developers building on LoopEngine will find the full reference in the docs.