Integrations
Integration guides for retrieving curated, sourced claim records and reviewing their evidence before an LLM assertion is used. Treat framework samples as starting points: pin current dependency versions and test the review boundary in your own application.
LangChain
Retrieve-then-cite + generate-then-review patterns. Python + JS examples for retrieving VERITAS candidate records and comparing cited evidence.
LlamaIndex
Custom retriever wrapping VERITAS search plus candidate annotation. Compatible with QueryEngine and ChatEngine; evidence review remains separate.
Haystack
Two Haystack 2.x components: a candidate-record retriever and an evidence-review stage. A bestMatch is similarity, not a truth verdict.
LangGraph
A StateGraph with candidate retrieval and evidence-review nodes. Compare generated assertions with returned statements and sources before assigning support.
OpenAI Tool Calls
Expose catalog search and candidate retrieval as native OpenAI function calls, with exact-statement and evidence comparison before citation.
Vercel AI SDK
Wire VERITAS into Next.js + AI SDK chains for tool-based candidate retrieval and post-stream review. TypeScript-first.
DSPy
Custom dspy.Retrieve backed by the VERITAS catalog plus a candidate-review module. Optimize against labeled support, not candidate rate.
Pydantic AI
Type-safe candidate retrieval as a Pydantic AI tool. Schema validation catches shape errors; factual support remains a separate decision.
Anthropic SDK
Expose candidate retrieval as a Claude tool via the Anthropic SDK. Python and TypeScript loops plus an evidence-review prompt and application guardrail.
Instructor
Jason Liu's structured-output library. Pydantic models retrieve VERITAS candidates at parse time; schema retries remain separate from factual evidence review.
Need a framework that isn't listed? Tell us — the next guide is whichever framework gets the most requests this month.