Wavesense
**GPT-5 powered RTL design understanding through hypothesis and experimentation** WaveSense is an automated tool that uses GPT-5 to understand digital hardware designs by intelligently generating hypothesis, running simu
Core trick
GPT-5 proposes RTL coverpoints, generates testbench stimulus, runs simulation to capture waveforms, then updates its model of the design.
Description
**GPT-5 powered RTL design understanding through hypothesis and experimentation**
WaveSense is an automated tool that uses GPT-5 to understand digital hardware designs by intelligently generating hypothesis, running simulations, and analyzing waveforms to build comprehensive mental models. This tool iteratively runs the experiments until a much better understanding of how chips actually behave is achieved.
## How does it achieve this?
Given an RTL design,
1. Hypothesis: Come up with interesting coverpoints / stimuli
2. Experiment: Generate code and run simulation to capture waveforms for these
3. Analysis: Create a mental model of how the design behaves based on the generated artifacts
4. Repeat from step 1.
README excerpt
# Wavesense ## What it does? **GPT-5 powered RTL design understanding through hypothesis and experimentation** WaveSense is an automated tool that uses GPT-5 to understand digital hardware designs by intelligently generating hypothesis, running simulations, and analyzing waveforms to build comprehensive mental models. This tool iteratively runs the experiments until a much better understanding of how chips actually behave is achieved. ## How does it achieve this? <img src="./workflow.png" alt="Wavesense" width="600" height="600" style="max-width: 100%; height: auto;"> Given an RTL design, 1. Hypothesis: Come up with interesting coverpoints / stimuli 2. Experiment: Generate code and run simulation to capture waveforms for these 3. Analysis: Create a mental model of how the design behaves based on the generated artifacts 4. Repeat from step 1. ## Implementation details - Codex CLI as orchestrator agent with the following tools: - Regular bash commands - Verilator/VCS as simulator - Compiled design as the environment - Waveforms as the observations ## Usage - `OPENAI_API_KEY=<your-key> python src/wavesense.py path/to/design.sv path/to/filelist.f my_artifacts_folder` Some example designs and filelists are in the `example_rtl` folder. ## Requirements - Codex CLI ``` npm install -g @openai/codex ``` ~/.codex/config.toml ``` model = "gpt-5" # full-auto mode approval_policy = "on-request" sandbox_mode = "workspace-write" ``` - vcdvcd, vcdcat ``` pip install