Show and tell: Wakeel, an agentic self-healing layer for OpenLane/OpenROAD

[Rishikesavan0129]

I built Wakeel, an open-source closed-loop EDA orchestrator for OpenLane/OpenROAD.

Wakeel monitors actual EDA-tool failures, diagnoses them using a deterministic knowledge base, applies a correction, reruns the flow, and uses the EDA tools themselves as the final verification authority.

The architecture follows the same broad self-verifying agentic EDA principle that has recently been demonstrated industrially by Siemens and NVIDIA with their Fuse EDA AI Agent, presented at DAC 2026. The idea is that an AI/agentic layer can orchestrate EDA tools while deterministic EDA engines remain the source of truth for verification.

Wakeel is nowhere near the scale or capability of Fuse EDA, and I'm not presenting it as an equivalent system. It is a much smaller open-source project focused on making this kind of workflow accessible to OpenLane/OpenROAD users.

As a concrete example, Wakeel autonomously diagnosed and resolved an OpenROAD nwell.4 tap-cell geometry failure, dynamically applied the required correction, reran the flow, and achieved a DRC/LVS-clean result without human intervention.

I'm particularly interested in feedback from people familiar with OpenROAD/OpenLane internals, physical-design failure modes, EDA automation, or agentic EDA.

GitHub: