Your P&ID drawing, now data you can ask

SQL tables and a graph that people and AI agents can query. A vision-language model finds the symbols tile by tile; the connections come from traditional computer vision (thresholding and connected-component labeling, no model); small scripts validate the graph and export it. No GPU, no training.

npx skills add dzakybd/pid-parser

Two versions: a skill for coding agents such as Claude Code and Codex, and a LangChain Deep Agents version for your own agent workflow. Source: github.com/dzakybd/pid-parser. How it works, step by step: the write-up. The sheet is one very dense P&ID (190 symbols, 732 connected symbol pairs); the two tabs are two runs on it. These results were computed ahead of time; this page calls no model.

sqlite3 pid.db < pid.sql          # or: psql -f pid.sql
cypher-shell -f graph.cypher       # Neo4j