CoLog is a new framework designed for detecting anomalies in operating system logs. It tackles the limitations of traditional unimodal and multimodal methods by using collaborative transformers to process various log modalities effectively. The framework incorporates a modality adaptation layer that harmonizes the different types of log data, allowing it to learn intricate patterns and relationships within the data. This approach significantly enhances anomaly detection capabilities. CoLog has been tested across seven benchmark datasets, including system logs from environments like Casper RW and Apache Hadoop. The results are impressive, with CoLog achieving a mean precision of 99.63%, mean recall of 99.59%, and mean F1 score of 99.61% for both point and collective anomalies. In contrast, traditional methods like logistic regression and support vector machines performed significantly worse, often with precision below 70%. The framework's architecture includes various components such as multi-head impressed attention and balancing layers, which optimize how different modalities influence the final detection outcomes. Installation of CoLog is straightforward, requiring Python 3.8 or higher and a CUDA-compatible GPU for training. Users can clone the repository, set up a virtual environment, and install necessary dependencies. There are clear commands for preprocessing datasets, training the model, and running tests, making it accessible for users with basic programming skills. The results from multiple datasets consistently showcase CoLog's superiority, marking a significant step forward in log-based anomaly detection for cybersecurity and system monitoring.