Moats Verify

Moats Verify checks AI-generated text against your document library and returns per-claim verdicts with citations.

What it does

Verify AI-generated text against your documents.

Works for RAG output, report fact-checking, compliance review, or any text you want to verify against source documents.

• Extracts factual claims from input text
• Retrieves candidate evidence from ChromaDB
• Re-ranks evidence with FlashRank
• Applies structured comparison (numeric, temporal, polarity)
• Falls back to LLM reasoning when structured comparison is inconclusive
• Returns SUPPORTED, PARTIALLY_SUPPORTED, CONTRADICTED, or NO_EVIDENCE

Requirements

• Docker Desktop (or Docker Engine + Compose plugin)
• An LLM provider key (OpenRouter/OpenAI/Anthropic) or a local OpenAI-compatible endpoint

Quick start (Docker, no local pip)

cp .env.example .env && docker compose up --build

Then open:

• App UI: http://localhost:3000
• API docs: http://localhost:8000/docs

Note: the Compose stack also starts a Neo4j container for backward compatibility. The active verification pipeline uses SQLite + ChromaDB.

First run checklist

1. Open Settings and configure provider, chat model, and embedding model.
2. Open Library and create a library.
3. Add documents via upload or local folder source.
4. Run Sync/Build to ingest and embed documents.
5. Open Verify and submit text.

Verify API example

curl -X POST http://localhost:8000/api/verify \
  -H "Content-Type: application/json" \
  -d '{"text":"Revenue was $5M in Q3 2024"}'

Response includes:

• verdict
• confidence
• reason
• evidence (text, source, page)
• used_llm (true = reasoning fallback, false = structured comparison)

Data and privacy

• Runtime data is stored locally under ./data/ and Docker volumes.
• User document folders are mounted read-only from ./library/.
• .env, library/, and data/ are git-ignored.

Development (without Docker)

pip install -r requirements.txt
uvicorn backend.main:app --reload --host 0.0.0.0 --port 8000

In another shell:

cd frontend && npm install && npm run dev

Project layout

• backend/: FastAPI API, ingestion, retrieval, verification pipeline
• frontend/: Next.js UI
• tests/: backend tests
• backend/deprecated/: legacy verify stack kept for rollback

Troubleshooting

• Empty verification results: confirm library has indexed chunks.
• NO_EVIDENCE for known claims: confirm the relevant documents are in the active library.
• LLM errors: validate provider settings on Settings page.

License

This project is licensed under the MIT License. See LICENSE.