Critical infrastructure, defense networks, and healthcare systems often operate under strict Air-gapped requirements where external network access is completely prohibited. Utilizing hosted language models in these environments is impossible due to strict zero-egress policies. Palimpsests delivers an ideal solution through an open-source inference engine designed from the ground up for isolated networks. Operating with no network egress, the system ensures that AI processing, memory state persistence, and audit logging remain entirely on your local, controlled hardware.
The system is optimized for modest local hardware, preserving system resources specifically for workload execution. Under operational pressure, its logging framework prioritizes system responsiveness, recording events as hash-chained entries without stalling inference loops. The engine uses a PALA-1 v1.0 append-only log format, enabling security personnel and auditors to validate system operations completely offline without requiring decryption keys or external witness servers. For organizations constrained by zero-trust or disconnected network mandates, Palimpsests brings dependable, highly efficient AI capabilities directly to the edge.