Skip to content

LongTracer

LongTracer

RAG hallucination detection, multi-project tracing, and pluggable backends — all batteries included.

PyPI CI Python License

What is LongTracer?

LongTracer is an open-source Python SDK that detects hallucinations in LLM-generated responses. It verifies every claim in an LLM output against your source documents using a two-stage hybrid pipeline:

  1. STS (Semantic Textual Similarity) — fast bi-encoder finds the best-matching source sentence for each claim
  2. NLI (Natural Language Inference) — cross-encoder classifies entailment / contradiction / neutral

The result is a trust_score (0.0–1.0), a list of flagged claims, and a full trace of the verification pipeline.

Why LongTracer?

Problem LongTracer's answer
LLMs hallucinate facts not in your documents Detects contradictions at the claim level
Hard to debug which claim failed Full trace with per-claim evidence mapping
Tied to a specific vector store or LLM Works with any RAG framework — just strings in
Verification adds too much latency Parallel pipeline: relevance scoring runs alongside LLM generation
Need to track verification across projects Multi-project tracing with pluggable storage backends

Install

pip install longtracer

30-Second Example

from longtracer import CitationVerifier

verifier = CitationVerifier()
result = verifier.verify_parallel(
    response="The Eiffel Tower is 330 meters tall and located in Berlin.",
    sources=["The Eiffel Tower is a wrought-iron lattice tower in Paris, France. It is 330 metres tall."]
)

print(result.trust_score)         # 0.5
print(result.hallucination_count) # 1  ("Berlin" contradicts "Paris")
print(result.all_supported)       # False

No vector store dependency. No LLM dependency. Just strings in, verification out.

Next Steps