This presentation introduces LogLMs, transformer-based foundation models specifically pre-trained on log sequences. LogLMs understand the 'language' of logs, enabling it to identify anomalies and deviations from normal behavior across diverse protocols and usage patterns. Unlike rules-based systems, LogLMs adapt to changing environments through active learning and federated fine-tuning. This approach provides holistic security, including anomaly detection, threat hunting, real-time alerts, compliance, and forensics. We will see how a LogLM, deployed as TEMPO, detects novel attacks, and empowers forensic analysis.