ORCA van Loon uses TypingMind to streamline processes

We use TypingMind for specialized agents across different areas, both internally (e.g., onboarding) and for recurrent editorial tasks requiring a specific tone of voice. The agents are trained with briefing data tailored to certain writing styles, enabling them to generate content that we later refine.

ORCA van Loon's chat portal using TypingMind

I also personally use it for small coding tasks, as I handle website development and maintenance but am not a programmer. This allows me to create custom WordPress modules and similar features that would otherwise require a developer.

ORCA van Loon's AI agent in action

ORCA van Loon's knowledge base in action

The AI agents are the most valuable feature, especially the ability to share them within the team or build personal ones. Additionally, the option to switch between different LLMs based on the task is extremely useful. The Knowledge Base is also significant, as it allows us to provide structured and extensive training documents. As a small agency, setting up our own vector database isn’t currently appealing, so we happily utilize TypingMind’s Knowledge Base instead.

Yes, we enjoy working with TypingMind. While the interface feels a bit less polished and "raw" compared to tools like ChatGPT, it’s straightforward enough for our team members to use, even if they aren’t particularly tech-savvy. Many experiment with it and create their own agents for specific use cases. We conducted one or two training sessions to reduce hesitations and showcase the possibilities.

We currently don’t have metrics to track how much more efficient work has become with TypingMind, as we need to use it longer. After about a year, we’ll be able to measure if less time is spent overall on comparable tasks.

While we lack solid data, the general feedback (after using TypingMind) is positive—repetitive, time-consuming tasks are completed much faster, allowing consultants to focus more on strategic and substantive work.

ORCA van Loon's AI metrics

I discovered TypingMind through YouTube while looking for ways to reduce AI tool costs privately. Several videos helped me understand its functionality, so I purchased the standard TypingMind version for personal use and was quickly convinced. This led me to the Custom variant for teams. I set up a test version, which we trialed internally with a small group.

I explored other tools but found nothing comparable. Other options had overly technical interfaces or required installation via GitHub, which isn’t my expertise and takes too much time. A ready-to-use software solution like TypingMind, where I only needed to add API keys, was the perfect fit.

I really appreciate the speed at which the tool is being developed and how quickly access to the latest LLMs is provided—huge praise for that! In terms of improvements, I would (also privately) love to see the integration of ChatGPT’s Realtime-API. Perhaps with its own basic user interface, not in the "Chat" section but as an additional point like "Voice" or something similar. Currently, I have to use OpenAI’s Playground for high-quality voice-to-voice conversations, which means I barely use this feature anymore, even though I really like it.

It would also be great if some of the interactive elements, now standard in ChatGPT, Perplexity, etc. (like Google Maps), could find their way into TypingMind.

We’ve also noticed, for example, that citations can sometimes be a problem. They are displayed differently each time, and sometimes they don’t appear at all. I don’t know if TypingMind itself can improve this or if it depends solely on the LLMs.

It would also be fantastic to further develop the Stable Diffusion plugin or add image generation as a standalone tool. The plugin’s current configuration options are very limited, and it would be incredibly helpful to have something like a settings navigation for image creation, similar to what platforms like MidJourney now integrate in their user interfaces. We use image generation a lot for inspiration, for initial briefings to graphic designers, and for presentations.

One unique feature could be the ability for multiple people to share an agent "live". This means several team members could literally see in real-time what’s happening in the chat and what responses or images the AI is generating. It could mostly work as a regular chat among team members on a specific topic where the AI becomes active or is called in by tagging it with @, and all members can follow along accordingly.

To improve cost management, it would also be helpful to show an actual cost estimate in the top-right corner, like in the private TypingMind version—not just an estimated amount of tokens. Many team members don’t understand or can’t interpret this "currency."

Additionally, it would help a lot to have a small info button behind each LLM, briefly explaining its strengths, weaknesses, and costs. I think most users of TypingMind for teams are not particularly tech-savvy. They just want to use an AI tool but might not know the differences between the LLMs and their optimal use cases.

A colleague also wishes it were possible to simply open individual chats as tabs so that multiple chats can be worked on simultaneously.

Lastly, a "Fact Check" function would be amazing. This could potentially be provided as a plugin when the major DeepSearch models are integrated. That way, you could work with a "normal" LLM and, for specific answers, press a Fact Check button to trigger a DeepSearch plugin to investigate the topic further across more sources.

The look of TypingMind itself doesn’t feel too modern and sometimes seems a bit cluttered or functional. It might benefit from a more visually appealing design.

Finally, it would be great if prompts could be imported more easily without needing to use a .jstor file.