Token cost per query — ckg-benchmark v0.6.2
11× fewer tokens · 4.7× better F1
RAG (retrieval-augmented generation) 2,982 tokens
CKG — context router over the AI documentation stack 269 tokens
A context router traverses declared edges. The graph already knows that TensorRT-LLM depends on CUDA — it routes your agent directly to the subgraph, skipping retrieval entirely. Benchmark: ckg-benchmark v0.6.2 · 28 queries · F1 0.471 vs RAG 0.123.