Monsel Agent Orchestrator
The Agent Orchestrator is where you set up, run and supervise agents that do real work: raise the invoice, draft the quote, sort the inbox. You decide what each one does, what starts it, what it may use and what it may read. Work you choose waits for a person before it goes out.
The anatomy
An agent is a colleague you set up for one job. Its instructions are a skill, written the way you would brief a colleague. Everything else is a setting you chose.
When an agent surprises you, its instructions, tools and run history give you what you need to find out why. The person who knows the business can change it by editing words.
What starts an agent
One agent can have several starting points.
A chat window on your site, your WhatsApp Business number or a Telegram bot. The agent replies where the customer started.
An address the agent watches. It replies in the same thread, with any files attached.
Every weekday at eight, the first of the month, once an hour.
Your website or shop sends Monsel a new order, and the agent starts with the order in front of it. How website messages reach an agent
A meeting starts or ends, so notes and follow-ups happen without anyone asking.
One agent announces an invoice was raised. The agent that sends invoices picks it up.
Chat agents can also act when a visitor goes quiet or closes the chat.
Connections
Connect an account once and choose which agents may use it. The password stays with the connection, never in the instructions.
Dropbox and OneDrive are coming.
Many business tools now offer one. Add it and its actions appear in the agent's list. How these connections work
Agents can use your own database without anything being opened to the internet. More on private networks
Point an agent at a model running in your own network with Ollama, so its thinking happens on your machine. How it works
Connecting an inbox does not let every agent send from it. You see exactly which tools each agent has.
Evaluations
Instructions are written and checked in Monsel Studio. Once real work starts, the Agent Orchestrator watches how they behave. The Agent Evaluator, the Skill Health Monitor, learning from your corrections and trying a draft skill against past conversations are available on the Pro plan and above. See pricing. Agent health is on every plan.
Nothing in this loop
changes an agent on its own.
Reads past runs and reports what went wrong and which instruction caused it. It never changes the agent.
When the same fault keeps coming back, it drafts a small change to the instructions for you to approve.
Each time you send work back with a reason, it is kept. When several say the same thing, a change is drafted.
One row per agent shows how often it hands conversations to a person, and flags a sharp change.
Approval of work
A reply to a customer, an invoice, a plan the next agent will act on. You see the draft, what the agent read and what it did. Then you decide.
A refund above a set amount, an unusual case, a document leaving the business.
Name the agent that carries on when you approve, and the one told when you send it back.
The reason is required, recorded on the document and given to the agent.
This is separate from live handoff, below. See the approval step in the walkthrough
Live handoff
The customer is upset, asks for a human, or wants something the agent was not built for. The agent says a colleague is coming. Whoever is on shift takes over, with the whole conversation in front of them. The customer stays where they started.
Webchat, WhatsApp and Telegram handoffs land in one place.
A chime on screen, and a phone notification when nobody is at the desk.
Close the conversation, or give it back to the agent to carry on.
Agents working together
Invoice agent is set to require approval. Front desk raises a request instead of calling it, and the call runs only if a person approves.
The Agents Map shows every agent, where a person approves and who passes work to whom.
Open an agent on the map to see what it can read and use, and any tool it used that is no longer allowed.
Runs and spend
Open a run to see what started it, what it read, each action and where it stopped. A failure shows the step that failed.
Co-Pilot
Co-Pilot sits inside the Agent Orchestrator and knows how it fits together. You can also draft agents through Co-Pilot.
Ask how anything works and it reads the product manual first. If the manual does not cover it, it says so.
Ask why an agent handed three conversations to a person yesterday and it reads those runs.
Ask how to set something up and it takes you to the screen where you do it.
On your phone
Monsel Agent Orchestrator installs from your browser. It opens in its own window with its own icon, and there is nothing to download from an app store. Open it once you have a workspace.
In Safari, tap Share, then Add to Home Screen.
In Chrome, open the menu, then choose Install app.
In Chrome or Edge, use the install icon at the right of the address bar.
Notifications can be switched on once it is installed.
Private beta
Monsel is in private beta with a small number of businesses. Leave an email and we will tell you when the next workspaces open.
To talk first, ask us to set agents up for your business from a written proposal, or write to hello@monsel.ai.
Analytics
With your permission, we use privacy-friendly analytics to see which pages are read. It sets no cookies and does not follow you across sites. We use no advertising and no cross-site tracking. Privacy policy