Collect data from several systems
The agent retrieves available information and prepares an overview. The answer includes sources and identifies data it could not obtain.
Output: a summary for reviewAI AGENTS FOR WORK TASKS
AIRun
AI agent platform for business
Create agents that collect data, prepare summaries and perform permitted actions in company systems. Start tasks in chat, on a schedule or from an event, then review the result alongside its action history.
AR / AGENT WORKSPACEHow AIRun executes a task
Select a node to see details
WHERE TASKS START
MODEL AND SYSTEMS
Direct model API
Through AIGateDirect model API
AR
The agent works on your task
Select a node to see details
AIRun starts the selected agent version with the required model and tools. Steps, errors and usage are saved; the run can be limited or stopped.
A model can connect directly without AIGate checks. Tool access goes through AIGate on both routes. DIAGRAM ELEMENT ↗
You set the systems, actions and limits available to an agent. Every run is saved in the history.
FROM SETUP TO RESULT
How to work with an AI agent in AIRun
Set up an agent once and use it for both one-off and recurring tasks.
Create or choose an agent
Set its instructions, model and permitted tools. Save an agent version or choose a published agent from the catalog.
Assign a task
Write in chat, fill in a form or send a task through the API. For recurring tasks, set up a schedule or an event trigger.
Follow execution
See the current step, requests to connected systems and model usage. Set limits and stop a run when needed.
Check the answer
Get a summary, list or table. Compare its conclusions with the sources and history: unavailable data and errors should be visible in the result.
RESULT
Repeat the task and verify the result using its saved history.
EXAMPLE USE CASES
What tasks can you give to AI agents?
Choose a workflow with known data sources and a clearly useful result.
Prepare a project and team overview
For example, compare project metrics from DevScore with team-level data from EmpScore. Connections and the data scope are checked during the pilot.
Output: metrics and explanationsRepeat a check on a schedule
The agent checks selected sources regularly or when another system sends an event. Rules define how repeated events and overlapping tasks are handled.
Output: changes over a periodMake a change with approval
When a rule requires approval, the agent waits for a person to decide before changing data in a system. Rejections and expired requests are saved in the history.
The first check uses a test objectFor each workflow, we connect the required systems and grant the agent access to specific data and actions in advance.
ACCESS AND SECURITY
How to control AI agents at work
AIRun manages tasks and history. A connected AIGate checks access to models and tools against configured rules.
AIRun / EXECUTION
- Agent creation and versions
- One-off and recurring tasks
- Limits and stop control
- History and results
AIGate / POLICIES
- Model request checks
- Tool permissions
- Approvals according to rules
- History of allowed and rejected calls
A separate environment for each task+
In Kubernetes, every run receives a separate Job. Long-lived keys remain on the server. Network restrictions are checked during installation.
Access for each role+
Users see their own runs. Security observers can review all runs without changing data. Company sign-in through OIDC or LDAP is configured during setup.
Connect company systems+
API and MCP provide interfaces for data exchange and tool calls. Nextcloud / WebDAV, SMTP email and webhook events are supported after configuration and testing.
Two ways to connect a model+
Connect a model through AIGate or directly through a compatible API. With a direct connection, AIGate checks do not apply to model requests. Tool access goes through AIGate / MCP Hub in both cases.
THE FIRST PILOT
Start with one work task
Use test tasks to check answer quality, access, limits and execution time. Then decide where to use the agent next.
Define a useful result
Choose a workflow and its owner. Prepare example tasks and agree on an answer that is suitable for work.
Connect data and a model
Grant the required permissions. Start with reading data and test changes in a system separately using a test object.
Check execution and errors
See what happens when data is unavailable, access is denied or an approval is rejected. Check limits, stopping and repeated events.
Assess the pilot outcome
Compare with manual work: time to a verified answer, human involvement, accepted results and model usage.
Pilot timing and costs depend on your task, sources, model and expected workload.
ANSWERS WITHOUT SMALL PRINT
Questions about AIRun
What is AIRun and who is it for?+
AIRun is a platform for creating and running AI agents for business tasks. Teams use agents to collect data, prepare summaries and perform permitted actions in company systems, with a history of each run.
How does AIRun differ from an AI chat?+
In AIRun, an agent has saved instructions, a model and available tools. Tasks start in chat, on a schedule or from an event. Each run retains the agent version, actions, errors and model usage.
Which company systems can be connected?+
Systems with an available API or MCP tool. During the pilot, we check compatibility, grant the required permissions and agree on permitted actions. Configurable connections include Nextcloud / WebDAV, SMTP email and company sign-in through OIDC / LDAP.
Can AIRun run in our infrastructure?+
Kubernetes deployment is supported, with a separate Job for each run. Resources, networking, storage, model access and operational requirements are scoped for the expected workload.
Can a stopped run undo an external action?+
A run can be stopped. A completed write in an external system is not automatically undone; the workflow needs its own recovery procedure.
Does an agent guarantee a correct result?+
Quality depends on the model, data and task. Acceptance criteria, limitations and execution history support human review before results are used.