We provide a generous AI budget, effectively unlimited for your normal workflow needs. You could use up to 10 runs per document to create the desired output. As a fair use safeguard, if consumption unexpectedly spikes over an average of 10 runs per document, we'll simply connect to offer a workshop and/or discuss it.

Of course, AI budget used during the development and testing of new workflows is covered. We strategically choose AI models: balancing cost-effective options for standard tasks with the newest, most powerful models for critical functions.

Our goal remains transparent pricing, aiming to deliver up to 70% cost and 90% time savings for you.

Prompt chaining is a method to enhance the quality of the final outcome in AI document generation. It involves using the output from one prompt as the input for another. This approach is particularly useful when you need to refine the end result.

For instance, instead of using a single prompt like "Provide an answer to the question, ensuring it follows the conversation flow," you could use two separate prompts. First, "Provide an answer to the question," and then, "Given this answer, rewrite it so it fits seamlessly into the conversation." This two-step process can lead to more coherent and contextually appropriate results, especially with complex and long documents.

But we don't stop at two steps, we build as many steps as needed for the perfect result.

These AI blocks are designed to improve their output quality through the process of manually labeling results. This labeling helps the AI learn and adapt for better performance.

An LLM, or Large Language Model, is a type of artificial intelligence model designed to understand, generate, and manipulate human language. These models are "large" both in terms of the size of the neural network (often consisting of billions of parameters) and the vast amount of text data they are trained on.

After 365 days, and/or whenever your storage capacity of 5GB has been reached, we remove run-data following the FIFO principle. Rest assured, your labeled data remains untouched, facilitating your experiments with various self-learning AI models.

Only authorized Twin personnel with the necessary permissions can access your data, and this access is strictly limited to maintenance and support purposes. We enforce robust access controls to ensure the security and confidentiality of your data. Your organization retains full control and ownership, encompassing all workflows and data transmitted through our system.

Yes, ensuring your data safety is a top priority at Twin AI. We are fully GDPR compliant and have implemented robust security measures. Our infrastructure is hosted within the EEA (EU), and we regularly have our security tested by third parties. Importantly, unlike generic AI tools, we do not share any data hosted in Twin AI to train external AI models; your data is only used to deliver and improve the services you request.
‍
Next to that: we are actively working towards industry certifications to further demonstrate our commitment to security. We aim to achieve ISO/IEC 27001:2022 certification by Q3 2025  and SOC2 certification by Q2 2026. For more information about our data security, we refer to our Trust Center here.

Essentially, you can use almost any type of information you normally would to create a document. The specifics depend on what you're building, but our system is designed to handle a wide variety, including company slide decks, notes from meetings or intakes, content from websites, PDFs, Excel files, and more. If you typically use it to write your documents, chances are our workflows can process it.

While tools like ChatGPT excel at general conversation and content creation, Twin AI is specifically built for automating structured business workflows. Instead of relying on single prompts, you gain control by building customized processes using interconnected blocks. This approach makes your AI workflows flexible, testable, and ensures your data is managed effectively.

Twin AI offers greater model flexibility than typical agent setups, letting you leverage a range of leading AI models. You can also model complex business decision trees and utilize advanced functions like data fine-tuning (with your explicit consent ) or embedding-based similarity checks. Ultimately, we provide a robust, controllable environment designed for specific business automation, not just general AI tasks.