Multi-Agent Workflow: Your AI Agent Team

On cloud.smartragents.ai you pick a lead agent and add contributors. One request, several AI agents, one answer structured by role.

Anyone working with AI agents knows the ceiling. One request, one agent, one answer. As soon as a task touches several disciplines, you had to run several chats. And stitch the results together yourself.

That ceiling is gone as of today. The multi-agent workflow is live on cloud.smartragents.ai. It arrives with three companions that bring order to daily work with AI agents: projects, secrets, and an open view of your agents’ plans.

The agent circle: your AI agent team at a glance

Agent selection has been redesigned. Instead of a list of identical cards, the eight specialist agents stand in a circle around the CEO. Like a team around its lead.

Each card shows a short name, an icon, a headline and an explanation. The CEO leads and summarises. The specialists bring their strength. On smaller windows it becomes a clean tile grid.

Multi-agent workflow: one request, several specialists

The heart of the update. You pick a lead agent, the CEO for example. Then you tick the contributors the task needs. A workflow bar appears above the input field. It shows you who is involved at all times.

Your first message gets an editable header carrying the brief for the team. Then the important part happens: the lead agent takes the request around the circle. Each contributor works on the part that belongs to their discipline. In the end you get one answer, structured by role.

Important for your budget: one workflow is one request. Three agents working on it does not mean triple the cost.

This opens up ways of working that used to be manual labour. A product text gets written by the content side, checked for message and tone by the CEO, and proofread for technical accuracy by the IT side. All in a single pass.

Projects: structure that grows with you

The chat list now has two levels. Projects sit as collapsible headings with their chats indented below. Pinned chats stay at the top within their own group.

You create a project with the folder icon next to the “New chat” button. You can rename a project or dissolve it. Any chat can be moved into another project with the folder icon. Dissolving deletes nothing, chats and folders stay.

Every project automatically gets its own output folder for your agents’ results. Behind the scenes, creating a project also creates a project of the same name inside your agent container, with its own instructions, skills and secrets. If you run several clients or ventures in parallel, you keep the overview. Without folder acrobatics in the file system.

Secrets: use credentials safely inside the chat

Under Memory you now find the Secrets tab. You store a value once, a password or an interface key for example. In the chat you then use it through a placeholder, such as §§secret(MY_PASSWORD).

The value is inserted only immediately before the tool runs. It never appears in the response stream.

One deliberate restriction makes this trustworthy: once stored, a value can never be read back. Not through the interface, not through the chat. If you want to change a value, you overwrite it.

You can now see what your agents are planning

The tasks tab shows a new block: “Planned by your agents”. Whatever your agents worked out in chat as ideas and next steps is now laid out in front of you. With a run button and a delete button.

After every agent run, the app refreshes memories, secrets and tasks by itself, even with the window open. You no longer have to go looking for what your team is up to. You can see it.

Under the hood: one watchdog for many customers

Something moved under the hood today as well. The SMarTrTrader watchdog is now multi-tenant and runs right next to the trading systems.

One watchdog looks after many customers. Each customer gets their own alerts, in their own language, in their own Telegram chat. A connection problem at one customer does not touch the others. Credentials are stored per customer and only where they are needed.

The basic watch costs nothing. Only real events are billed, against a credit balance per customer. The watchdog never trades by itself. It observes and it reports.

Preview: SMarTrChrome learns to pause

We are working on the next update for the browser agent SMarTrChrome. It has one theme: pausing.

In future, the autonomous browser agent will stop before sensitive steps. Before accounts, keys, two-factor codes and permission changes. It then shows you an approval card. Your yes applies to exactly that one step, matched character by character against what you saw.

You stay in control, even while the agent does the work. We will report back as soon as the release is in the browser store.

Frequently asked questions

What is a multi-agent workflow?

A multi-agent workflow is a single request that a lead agent takes through several specialist agents. Each agent handles the part that belongs to its discipline. You get one answer back, structured by role.

Does a workflow with several agents cost more?

No. A workflow counts as one request, no matter how many agents contribute.

How many AI agents can I brief at once?

You pick one lead agent and add as many of the eight specialist agents as contributors as you like.

How do I pass a password to an AI agent safely?

Through the Secrets tab. You store the value once and call it in chat through a placeholder. The value is inserted only at run time and can never be read back afterwards.

What are projects for?

Projects group your chats, give every venture its own output folder, and create a project of the same name in your agent container with its own instructions, skills and secrets.

Conclusion

Multi-agent workflows, projects, secrets and visible plans. Today your cloud went from being a tool to being a team lead. You make one request, a team of AI agents works it through, and you keep structure, budget and control.

Human with machine, every day.

Try it: cloud.smartragents.ai, your AI team is already waiting.