Run agents
For each agent, the task, permitted tools, limits and processing status are configured in Studio.
Swiss AI Agents: carrying out tasks, with approvals.
The agent layer of the platform: agents receive a task, use approved tools, work through intermediate steps and request approvals at defined points. Several agents can work on subtasks.
In brief: A chatbot answers questions. A fixed workflow carries out defined steps. An agent handles a clearly scoped task, uses permitted systems and asks a person at defined points. Swiss AI Agents is the execution layer for this.
Tasks
What the agent layer does.
Take on tasks
Triage an email, check a document, prepare an appointment, create a quote, compile a report.
Use tools
ERP, CRM, DMS, calendar, telephony, search or knowledge systems as approved.
Start workflows
An agent can start, pass on or complete a defined process.
Process triggers
An incoming email, call, document, appointment or schedule can start an agent.
Coordinate several agents
One agent breaks the task down, others handle subtasks, one agent brings the results together.
Approval by a person
Defined approval points before sending, booking or committing.
Processing status
Open, pending or done – documented in Studio to the agreed extent.
Example
An agent handles an enquiry.
Which products are involved in the example depends on the project – see the overview in the products hub.
| Step | What the agent does | Approval |
|---|---|---|
| Receipt | Recognise the email, classify the request, look up the customer in the shared data core | – |
| Checking | Spot missing details, prepare a follow-up question | – |
| Preparation | Create a draft quote in Easy Factura, assign project documents | – |
| Response | Draft a reply | A person checks and sends |
| Completion | Update the CRM, close the task | – |
Limits
Agents do not make decisions on their own about clients, money or contracts.
For each agent, we define permitted actions, data access, processing states and approval steps. Impermissible actions are to be blocked and logged. Limits and error behaviour are tested during the trial run.
Swiss AI Factory decides which models an agent uses, based on capability, data protection, cost and availability.
Frequently asked questions
Frequently asked questions about Swiss AI Agents
How many agents run in parallel?
That depends on the tasks, models and compute nodes. We check specific capacities for each project.
Do you call it an agent farm?
Publicly, we talk about the agent layer or agent runtime – that describes the function more precisely.
Can an agent run locally?
Local operation on your own servers is possible. We check the suitability of compute nodes, interfaces and operating model in the project.
Do you show internal details?
Publicly, we show the principle, layers and data flow at a high level. Internal topology, routing rules, security rules and version states stay internal.
Is it already in production?
We state the maturity of each building block openly. Phrases such as “is designed to” describe the architecture, not a client deployment. Studio, Agents and parts of Factory run in our own development and operating environment and are continuously extended. Further operating models, multiple nodes and private and local installations will follow step by step.
Next step
The task you want an agent to take on.
We check the task, tools, approvals and operation against your process.