Add Paralleliq — model-aware GPU control plane for AI inference clusters

[paralleliq]

Paralleliq is a model-aware GPU control plane that sits above inference clusters and understands both the infrastructure and the models running on it.

Most tools in the Inference Platform section handle how models are served. Paralleliq handles where models are placed and whether they are placed correctly. It detects four patterns that cause 20-40% GPU spend overhead in production inference:

• Tier misplacement - a 7B model running on an H100 that only needs an A10G
• Dark capacity - GPUs allocated to deployments with no live traffic
• OOM risk - models approaching their GPU memory ceiling
• CPU:GPU imbalance - CPU saturation throttling GPU throughput in agentic workloads

Each finding is quantified in dollars and delivered with human-in-the-loop approval workflows and a full audit trail. Also ships piqc (https://github.com/paralleliq/piqc), an open-source read-only scanner for quick fleet audits.

Added to Inference Platform as the management/control layer for teams running these systems at scale.