Build with Amistio: create, try, and inspect an agent
Make an agent, test it safely, and read the record of what it did — from first draft to a version people can run.
By the end of this page you can make an agent, test it safely, and read the record of what it did — with no surprises when you turn it on.
What you need
- An Amistio account.
- A desktop web browser — building is a desktop activity. A phone browser shows the same Studio, but editing is cramped.
1. Start in the Studio
Building starts on My agents — amistio.com/builder simply brings you there. Every agent you own has a quiet Edit link on its card, and its More menu holds Archive and Delete; archived agents wait in an Archived fold at the bottom of the page, where Restore brings one back. Once you have six or more agents, a search box and a status filter appear above the cards. Press Edit to open the Studio for that agent: it is a focused editor without the left column, and its My agents link at the top brings you back. The account button at the bottom of the left column holds Settings, Skills, Dark mode, and Help & guides.
2. Create an agent
Press New agent — the green button in the left column. It opens your Assistant: say what the agent should do in plain words, and your Assistant drafts it for you. The draft shows up as a card in the chat, with buttons to turn it on, change it, or open it — your Assistant never turns an agent on by itself. To start from a ready-made agent instead, open Explore.
3. Read the canvas like a sentence
START WHEN (manual, schedule, Telegram, RSS, webhook), THINK WITH AI (the AI step), ACT (actions and Connected apps). A from-scratch draft starts with a manual trigger, an AI step, and a result.
4. Configure, then try it
Give your AI step real instructions, then press Try it to run the draft you are editing. Nothing is turned on by this; people with your share link only ever run the saved copy that is on.
5. Read the outcome honestly
A green SUCCEEDED badge means the run completed — not that it did meaningful work. If your draft's steps have no instructions yet, the tester says so in plain words right next to the result and points you at the AI step, so an empty success never fools you.
6. Inspect the record
Open the Runs tab to inspect every test and real run of this agent, step by step.
7. Turn it on when it behaves
Use Turn on & share to save an exact copy people can run. Share links always run the copy that is on, never your draft — and sharing never copies your connected accounts.
Working with other people? Use a room
Rooms are shared spaces where several people put agents to work together. They live inside Conversations, not the Studio: open Conversations, press New room under New chat, and you are in the room. Its Details is where you invite people with a link, add or remove agents, and rename or archive the room. Everyone who joins finds the room in their own Conversations list.
Connecting your own software? You are already in the right place
Developers are builders in Amistio. Everything for connecting your own systems — accounts, connectors, webhook starts, and MCP tools — lives under Connected apps in the left column; Skills live in your account menu. Open Connected apps and choose Developer guide in the page header to read the public Developers page, which maps exactly which connection paths exist today. There is no separate developer side inside the app.
Want the agent as a file? Export it, or import one
Every agent is also a small JSON file — the same file the API reads and writes. In the Studio's More menu, choose Export agent file to download it (it ends in .agent.json and never holds a password or key). To bring a file in, press New agent and choose Import an agent file: pick or paste the file, and Amistio runs the same checks it runs when you save by hand. If something is off, you see the same findings the Studio shows, nothing is saved, and you fix the file and try again.
What you should see
A draft agent whose test runs appear in its Runs tab, with results you can explain — and no surprises when you turn it on.
There is no all-agents run list today: run history lives with each agent, in that agent's own Runs tab.