AgentsOperational System Kernel for Agentic Runtime
Has many agents. But only one boss
OSKAR is the AI platform for your agents. In your own network or in the cloud.
It orchestrates AI agents, connects your systems and answers questions about the company, on a single Mac mini, on your own server farm or in the cloud. When OSKAR runs in your own network, your data does not leave the building.
In your own network or in the cloud

Six sentences
Almost every company wants to use AI
And almost everywhere the same concerns come up. These are the six sentences we hear most often.
“Our data must not leave the building.”
Then OSKAR runs on your own hardware with a local language model. Requests and answers stay in your network.
“Every team is building its own AI helper.”
OSKAR is the one platform for all agents in the company. Run the same way, controlled the same way.
“Agents acting without asking make us nervous.”
Every agent has a control level: Observe, Approve or Automatic. Anything with consequences, OSKAR puts up for approval first.
“Our knowledge is spread across ten different tools.”
OSKAR connects systems such as Salesforce, Notion, NEVA and TESS and brings the information together into one answer.
“We do not know whether our own hardware is worth it.”
Then start in the cloud or with a single Mac mini. As demand grows, more nodes are added.
“Afterwards nobody knows why the AI did that.”
For every action, OSKAR shows what happened, why, and what comes next.
How it works
From task to result
- 1
Task
A question in the EVI chat, a schedule or an event, such as an alert from TESS. OSKAR takes the task and works out what needs doing.
- 2
Orchestration
OSKAR hands the work to the right agents, pulls data from connected systems and NEVA, and computes on whichever node is free.
- 3
Result
The answer comes back with sources. Or, if a step has consequences, as a proposal that a person approves.
Agents
Agents are the specialists in OSKAR. Hermes finds funding programmes, Jev analyses content, others answer questions or chase late data sources. For every agent you decide what it may access and how independently it acts.
- Agents
- Hermes, Jev and your own for your workflows
- Access
- Set per agent
- Control
- Observe, Approve or Automatic
- Traceable
- What happened, why, and what comes next

Nodes
OSKAR runs on a single Mac mini just as well as on several servers. Every computer is a node, and OSKAR distributes the work itself: language models, analyses, transcription for the NEVA Recorder. The overview shows how many requests were handled in your own network and how many went outside.
- Nodes
- From Mac mini to server farm, in the cloud too
- Distribution
- Tasks run wherever capacity is free
- Models
- Local language models, search, transcription

Connections
OSKAR draws information from Salesforce, Notion, NEVA, TESS and other systems and returns answers through EVI. For every connection it is set whether OSKAR only reads or also writes. Approval rules decide when a person has to agree, for example before an agent changes something in Salesforce.
- Systems
- Salesforce, Notion, NEVA, TESS, EVI and more
- Permissions
- Read or write, per connection
- Approvals
- Rules for who has to confirm what

With EVI and NEVA
Together with EVI, OSKAR becomes someone the whole team can talk to. Whoever asks in the chat gets the answer from OSKAR, gathered from agents, NEVA and your systems. NEVA can also use OSKAR as its AI source. When OSKAR runs in your own network, all processing then stays there.
- With EVI
- Answers in chat and channels
- With NEVA
- Knowledge as a source, OSKAR as NEVA’s AI
- Independent
- OSKAR also runs without NEVA and EVI

Growth
OSKAR grows by adding nodes, in the office or in the cloud. Agents, connections and rules stay as they are
Getting started
One Mac mini
For the start and for smaller teams. Chat, questions about the company and the first agents run on a single device in the office.
Growing
Several nodes
More computers take on larger models, analyses or transcription for the NEVA Recorders. OSKAR distributes the work itself.
Full capacity
Your own server farm
For many users, large models and many agents at once. In-house or in the cloud.
Control
In your own network with a local language model, or in the cloud. You decide what OSKAR may do
Local AI
In your own network
In your own network, the language models run on your hardware. Requests, documents and answers do not leave your network.
Connections
Connected on purpose
You decide which systems OSKAR can reach. Each connection is either read only or read and write.
Control
Approval before action
Agents act only as independently as you allow. Steps with consequences need a person’s approval.
For NEVA and EVI
One AI for everything
NEVA and EVI can use OSKAR as their AI source. When OSKAR runs in your own network, their processing stays in-house too.
Working together
OSKAR works on its own. And even better with the other platforms
Questions
What people often ask
What hardware does OSKAR need?
None, if OSKAR runs in the cloud. In your own network, a single Mac mini is enough to start. As demand grows, more computers are added as nodes, up to your own server farm.
Does our data leave the network?
Not when OSKAR runs in your own network. The AI then works locally on your hardware, and requests and answers stay in-house. OSKAR only reaches external systems such as Salesforce or Notion if you connect them.
Which agents are there?
For example Hermes, who finds funding programmes, and Jev, who analyses content. On top of that come agents set up for your workflows.
May agents act on their own?
Only as far as you allow. Every agent has a control level: Observe, Approve or Automatic. OSKAR puts steps with consequences up for approval.
Do we need NEVA and EVI for this?
No. OSKAR runs independently. With NEVA, OSKAR gets access to your company knowledge, and with EVI an interface the whole team can ask through.
Many agents, one boss. In your own network or in the cloud

“Automation starts with a clear picture of the workflow. If you cannot model your process, you automate your mistakes.”
