- / MULTI-AGENT RAG
- / EVIDENCE VERIFICATION
- / TRUST SCORING
We build answers you can audit — grounded in your documents, scored for trust, and explained step by step.
Evidence. Trust.
Consensus.
TrustRAG doesn't just respond — it retrieves, verifies, scores, and shows its work.
Learn to see
brilliantly.
From the first upload to the final answer, TrustRAG turns raw documents into decisions your team can defend — quietly, precisely, at speed.
Grounded retrieval
Every claim is traced back to the exact chunk it came from, ranked by similarity and trust level.
Layered verification
A research agent drafts, a fact agent checks, and a trust agent scores before anything reaches you.
Consensus engine
Independent agents must agree. Disagreement is surfaced, never smoothed over.
Four steps from file to defensible answer.
Your documents stay in your configured TrustRAG storage. Retrieval is grounded in those documents, with optional Tavily verification for low-confidence answers.
Ingest
Drop in PDFs, DOCX, TXT, MD, CSV or JSON. The backend stores the original file, extracts text, chunks it, generates embeddings and marks it ready only after indexing succeeds.
Retrieve
Your question is embedded and searched against the authenticated user's tenant-scoped vector index. The strongest passages are passed to the verification agents as context.
Verify
The fact agent re-reads each retrieved passage against the draft answer, flagging unsupported statements before they are shown.
Score
Confidence, trust and consensus are computed per answer, with the evidence panel listing rank, similarity and source for every citation.
Four agents, one auditable answer.
Each agent has a single job and leaves a trace, so you can see exactly where an answer came from and where it was challenged.
Retrieval agent
Finds the passages that actually matter, filtered by the document scope you select.
Research agent
Drafts an extractive answer and maps every sentence back to a numbered citation.
Fact agent
Checks each claim against its evidence and downgrades anything thinly supported.
Consensus agent
Compares independent reads of the corpus and reports the level of agreement.
See the passage,
not just the claim.
Every citation opens the exact chunk it came from — with its document, rank, similarity score and trust level. Narrow retrieval to a subset of sources, then export the whole transcript with its evidence attached.
Sign in to inspect the real chunks, similarity scores, and trust signals produced from your own documents.
Built for work that has to hold up.
Anywhere an answer needs a source attached to it, TrustRAG gives you the paper trail by default.
Legal review
Search contracts and case files with a citation for every clause you rely on.
Clinical research
Interrogate protocols and papers without losing the provenance of a single figure.
Finance & policy
Turn filings, memos and regulation into answers your committee can defend.
Research teams
Build a shared corpus and let the whole team ask it questions in plain language.
The details, plainly.
Your files stay in the configured storage; answers are generated by the authenticated backend and AI service with inspectable evidence.
Where do my documents go?
Documents are processed by the configured TrustRAG services. Optional external verification can send the question to Tavily when enabled; set EXTERNAL_VERIFICATION_ENABLED=false to disable it.
What does the trust score actually mean?
It weights retrieval similarity against source authority and how many independent passages back the same claim.
Which file types are supported?
TXT, MD, CSV, JSON, PDF and DOCX are processed by the AI service. A document is not marked ready until text extraction, chunking and embedding complete.
Can I limit an answer to certain sources?
Yes. The search scope control in the chat lets you restrict retrieval to any subset of your indexed documents.
Upload a document.
Ask it anything.
Your corpus is indexed in seconds and every answer arrives with its evidence.