How it works
The first time you click the round button, Molted creates one instance for the agent and puts an API key of your account in its vault. The key is listed as "Management agent" in API Keys: revoke it there and the agent loses access. The agent reads the API reference before calling an endpoint, and works through the API exactly as the dashboard does. You can connect your Claude or ChatGPT account to it; without one, it uses the model included with Molted.
It asks for a clear yes before creating, resizing or starting an instance (that costs money from your balance), and before deleting anything. Reading, listing and diagnosing need no confirmation.
An agent for one of your clients
Say who the client is and what they do, for example: "Create an agent for my client Anna Miller, she runs Miller & Sons Realty in Austin." The agent tells you what it will create, waits for your yes, creates the instance, writes what it knows about Anna in her agent's profile, and gives you her sign-in link.
- What you do: confirm, then send the link to your client.
- The link opens her agent in Mue Web, or in your own app if you have one.
- A link given in the chat works once and for 10 minutes. For an e-mail, ask for an invitation: it works once and stays valid up to 48 hours.
Your branded app, on your domain
Ask: "Create the app for my clients, my website is example.com." The agent takes your logo and colors from your site, creates the application, and asks you to add one DNS record at your domain provider.
- What you do: add the CNAME record it gives you, for example app.example.com pointing to the address it names, then tell the agent it is done.
- Use a subdomain such as app.example.com: a bare domain cannot carry a CNAME.
- If your DNS is behind a proxy (Cloudflare's orange cloud), set the record to "DNS only" for the check.
- Once the record is visible, the agent checks it, the certificate is issued in about a minute, and it sends you the link to your app.
An instance that needs attention
Ask "What needs attention?". The agent reads the status, alerts and measurements of your instances and explains in plain words what is wrong, for example a storage that is 96% full. For a full disk it lists what takes the space and what can safely go (caches, build outputs, old downloads), with the size of each.
- What you do: say yes to the list it proposes. Nothing is deleted before that.
- The user's documents, the agent's workspace, memory and conversations are never removed without being named.
- In the dashboard, a storage at 90% or more shows as an orange ring around the green health dot.
A template from one of your agents
When an agent works well for one client, ask the management agent to turn it into a template. It reads that agent's files, removes everything private to the client (names, contacts, conversations, figures) and shows you the result before saving the template.
- What you do: read what it kept and removed, then confirm.
- New agents created from the template start with the same setup, without the first client's data.
Agents for a whole team, from one sentence
Give the agent a list of people, or ask it to get the list itself: "Email Jane, the HR manager at Northwind, and ask for the employees who need an agent. When she answers, create one for each from the sales-assistant template and email them a sign-in link valid 48 hours." It tells you how many instances it will create and what that costs, then creates one agent per person and prepares an invitation link for each, valid up to 48 hours.
- What you do: confirm the list and the cost, and approve the e-mail before it goes out.
- The e-mails are sent with your own connected mailbox: Molted sends nothing in your name.