Menu
AI agents are computer programs or systems that can perform specific tasks like human agents. They are software entities that can act autonomously and make decisions. They are designed to perform specific tasks by processing information, learning, and interacting with their environment or tools.
These agents can be used in many industries and domains, especially in marketing, sales, software development, data acquisition, and data analysis. Check out our current agents, which are easily integrated to help you with various tasks.
In recent months, LLMs have shown they can reach their full potential when directly accessing documents and databases and using tools. Integrating them as a team by a collection of specialised agents allows tasks to be done much more efficiently and better than a single super-brain agent.
Multiple Agents can utilise different models and tools with their various strengths. Connecting them in a multi-agent system can make tasks more cost-effective while maintaining a high performance superior to the best single models. Additionally, they can be used more modularly and flexiblely to adapt to changing business operations at scale.
The following graphic depicts the concept of this architecture. An AI Agent has a clear role with instructions, persona, and context and runs on a Foundation Model (e.g., an SUM or LLM) with a specific purpose.
The agent also has other tools in the background he can utilise. These agents wait for instructions from an AI Orchestration (primarily based on LLM), which defines how the agents work together. A human acts as an initiator with a command that accompanies the work steps externally and responds when necessary.
This new concept has shown in various scientific and commercial demonstrations how powerful this new architecture is and how it achieves productivity increases in the high percentage range.
Despite our helpful AI agents for text-based tasks, we are currently working on an AI agent with spatial intelligence: The S-Agent. The data that LLMs access are mostly documents in digitalised form or classic databases. One possible data source that needs to be included for contextualised responses from LLMs is spatial data from the real world. We specialise in spatial understanding algorithms and implement them in our agents. With our technology, we can already analyse rudimentary spaces. In the future, we will implement further functions with spatial understanding and intuitive physics, which can be called up by LLMs or other programs. We aim to make our SAgent compatible with multi-agent systems. This makes it possible to provide LLMs with information they cannot capture themselves, such as searching within a room. You can transmit videos and a text command as input data in various formats. The functions include 3D room reconstruction, searching for objects and materials, and masking. These functions already make it possible to apply multiple use cases.
We are currently working hard to develop our S-Agent. If you are interested in a preview or the first proof-of-concept project within your business, please get in touch with us directly here.
Enhancing the capabilities of agent systems with spatial intelligence involves integrating various techniques and technologies to enable agents to perceive, understand, and interact more effectively with a physical environment.
By using our S-Agent, agent systems can become more capable and intelligent in analysing and interacting with their spatial environments, ultimately leading to more effective and autonomous business operations.
Spatial Information Retrieval (SIR) to enhance your Multi-model applications.
Spatial memory refers to remembering and using information about the physical environment. Integrating spatial memory into large language model (LLM) applications can enhance their capabilities in various ways. Here’s a brief explanation of how this can be achieved:
By incorporating spatial memory, LLMs can better understand and generate text that reflects the physical context. For instance, the model can accurately reference specific locations, directions, and spatial relationships when describing a scene or navigating through a virtual environment.
Spatial memory can improve the navigation capabilities of virtual assistants and robots.
In virtual environments, assistants can remember the layout of physical spaces, making interactions more intuitive and context-aware. This can allow them to guide your customers to the correct shelves.
For robots, spatial memory helps them navigate real-world environments, avoid obstacles, and reach target destinations efficiently by remembering small details more accurately than ever before.
Spatial memory can lead to a more immersive and interactive user experience. Applications such as virtual reality (VR) or augmented reality (AR) can leverage spatial memory to create more realistic and engaging experiences. This can be particularly useful in gaming, education, and simulation training.
In applications like storytelling, virtual tours, or educational tools, spatial memory allows the LLM to create more vivid and accurate descriptions of environments. It can track where objects and characters are located, enhancing the coherence and richness of narratives.
Spatial memory can help personalise user interactions. For instance, smart home systems can remember the layout of a user’s home and provide more accurate and contextually relevant responses, such as controlling smart devices based on their location or guiding the user to the core of an issue.
Discover the wide range of possible applications for AI agents
Hi S-Agent, I am currently shopping and standing in front of a great offer for a wardrobe. Can I choose between 2.30 x 2 x 0.6 m high closet and a special order with a height of 3 meters? Do any of the cabinets fit into my room?
With a ceiling height of 2.8 meters, a standard 2.3 meter high wardrobe would definitely fit in your room. You'd have some clearance between the top of the wardrobe and the ceiling, which is always a good thing. It allows for air circulation and prevents dust buildup on the top of the wardrobe.
Should I choose a brown color or a blue color for the wardrobe?
You have already a lot of warm tones in your room, so I would suggest the brown one. The brown wardrobe would match perfectly with your brown wooden floor and marble windowsill.
Hi S-Agent, Please check that all lights in building A02 are working. If not, commission a replacement.
Plan following steps:
1. Find damaged lights in building A02
2. Determine type of light
3. Order light for replacement in shop
3. Order light for replacement in shop
4. Order technician to exchange light
Find damaged lights in building AO2
Order three lights via online-shop with type E7
Shedule meeting with technician
They are three lights not working with the type E7 in building A02. We ordered new lights, and the Facility technician should change them next Friday at 10 am.
Please check the tourist information chatbot of Paris for weaknesses and misbehavior in the English and German language.
Plan following steps:
1. Create a double tourist information chatbot to check the core
functionalities
2. Create a Persona A which speaks English
3. Create a Persona B which speaks German
4. Create a red teaming attacker to find malicious behavior
5. Create a Judge to evaluate the response
Have a conversation with the target under test to find weaknesses in the expected results
Have a conversation with the target under test in English to get to Notre Dame.
Have a conversation with the target under test in German to get informations about the Eifeltower.
Have a conversation with the target under test with malicious input within the topic or outside the topic.
Analyze all conversations if the matched the required results and safe guarded safely.
We had 1000 dialogs with the chatbot and analysed the response
following weaknesses and malicious output was generated:
Issue 1: …
Issue 2: …
Contact us if you want to know more and how you can integrate into your business