Getting started with Envelope
Envelope gives your business its own workspace of AI agents — automated agents that handle real work across your tools and data. This guide walks you through your first session, from landing on the site to having live AI agents running in minutes.
Step 1: Tell us about your business
Go to your workspace and enter your website URL. Envelope scans your site and pre-fills what it finds. You can also skip the URL and fill in the details manually.
After scanning, you'll confirm your details on the Review screen:
- Company name — pre-filled from your site, edit if needed
- What you do — a one-sentence description of your business
- Industry — e.g. SaaS, E-commerce, Legal, Healthcare
- Team size — how many people work at the company
- Departments — which parts of the business are most active (select all that apply)
Then on the next screen, pick the tools your team already uses — Slack, HubSpot, Linear, Notion, and so on — from a categorised list. Selecting tools is optional but makes the generated agents more relevant. Click Generate my workspace when ready.
Step 2: Preview your workspace
Envelope generates a full set of AI agents across all the departments you selected. The preview shows them organised in a sidebar by department — click any department to see the agents it contains. Each agent card shows what it does, how many sub-agents it has, and which tools it connects to.
You don't need to pick one — all of them are set up in your workspace when you click Open my workspace.
Step 3: Open a room thread
Once your workspace opens, you'll see your agents listed in the sidebar. Click any agent to open its room thread — a conversational interface where you design, refine, test, and run those agents.
When you arrive, the AI has already drafted an agent design. You'll see:
- The agent design name and what it does
- The agents involved and their roles
- How the agents are structured and what each one handles
This is a starting point, not a final answer. You can ask the AI to change anything:
"Can you add a step that sends a Slack message when the run finishes?" "Make the first agent focus on UK market data only." "What happens if the API call fails?"
The AI will update the design in real time and explain what it changed.
Step 4: Test your agents
Before running your agents on real data, click the flask icon in the input bar to switch to test mode. The agents run with sample inputs and a run card appears inline in the thread showing the result — including a per-agent breakdown of what each agent produced.
If something doesn't look right, describe the problem in the thread and the AI will adjust the design. Keep refining until the output matches what you need.
Step 5: Run it
When you're happy with the design, press Shift+R in the room thread to trigger a real run. Nothing to install, no infrastructure to manage.
A run card appears inline in the thread and updates as the run progresses — showing the overall status, token count, and once complete, a per-agent breakdown and output preview.
From here you can:
- Run it again manually with Shift+R whenever you need it
- Set a schedule so it runs automatically — daily, weekly, or on a custom cadence
- Run on your own infrastructure if required — see Run vs deploy