1.0 KiB
1.0 KiB
AgenticCode
Turns a large Natural/Java codebase into a queryable graph, so AI agents can answer code questions via API instead of reading source files.
Source Code → Parse → Graph (Neo4j) → Enrich → Agent API
What it extracts: call graph (CALLNAT/PERFORM), DB access (READ/FIND/
STORE + SQL), data structures (DEFINE DATA, PDA/LDA), variable-level
dataflow, control flow.
Why it's a great fit for agentic AI:
- Structured JSON answers instead of raw file text — no re-parsing needed
- Small, targeted lookups instead of holding the whole codebase in context
- Matches agent reasoning: ask one precise question at a time (callers → callees → DB tables → dataflow)
- Traces data lineage across programs — a mechanical task AI does reliably, humans do slowly
- Scales: 6,300+ files mapped in ~1 minute; queries return in seconds
Result: tasks like "what would it take to rewrite this program?" go from hours of manual tracing to one agent session with a complete, correct picture.