vust

See The Source · Cross-Check The Claim

See Where a Claim Comes From — Then Cross-Check It.

On the Research mode @vustSearchBot returns an answer with inline [N] markers and a Sources button — the links open, so you can read the evidence behind a sentence yourself. Want a gut-check on a contested number? The Second Opinion Council inside @vustbot asks three independent models the same question and flags where they disagree. Cardless, in Telegram.

Openable sources on Research · 3-model cross-check · in TelegramSources you can open · disagreement flagged
Openable sources on ResearchFlags where models disagreeCardless, in Telegram

Two surfaces, kept separate

Sources from Search — disagreement flags from the Council

These are two distinct tools, and we don't blur them. On the Research mode @vustSearchBot gives numbered markers and openable source links behind specific statements; Quick and Standard open the same Sources button, but nothing in the text points at a specific link. The Council of Sages inside @vustbot compares three whole answers (Claude, Grok, Gemini) and flags where they diverge — answer-level agree/disagree, with no web source links. Use Search when you need the evidence; use the Council when you want to know whether independent models actually agree.

Openable sources: @vustSearchBot on Research. Second opinion across 3 models: @vustbot → Council.
Specimens

See the difference

A source-free chat answer, versus a sourced Research answer plus a 3-model cross-check.

A plain AI chat answer

What you get

One fluent paragraph that sounds authoritative but has no links. You can't tell which sentence is grounded, which is a guess, or where a number came from.

The gap

To verify anything you leave the chat, re-search each claim by hand, and hope the model didn't invent a statistic. Trust is all-or-nothing.

@vustSearchBot in Research mode

You ask

"How much of global electricity came from solar in 2024, and is it growing faster than wind?"

You get

Several sub-queries are run and synthesized into one answer carrying inline [1][2] markers, with a Sources button that opens the matching list of links — so the sentence about solar's share points to a page you can read yourself. Quick and Standard skip the inline markers — the list still opens behind the Sources button, but nothing ties it to a specific sentence; pick Research when that link is the point.

Second Opinion cross-check (Council in @vustbot)

Same question, three models

The Council asks Claude, Grok (xAI) and Gemini the same question independently.

You get a verdict

An arbiter returns an AGREED section (all three concur solar grew fast) and a DISAGREED section flagging that they diverge on whether it outpaced wind in 2024 — so the contested number is marked, not buried.
Practical use cases

See the source, then check if the models agree

Researchers & fact-checkers
Need to know which sentence in an AI answer is actually grounded
In Research mode @vustSearchBot returns inline [N] markers tied to specific statements plus a Sources button of openable links — click straight through to the evidence behind a claim. Quick and Standard drop those inline markers, though their source list is still one tap away.
Anyone weighing a contested number
One model sounds confident, but you're not sure it's right
The Council of Sages inside @vustbot asks Claude, Grok and Gemini 3.1 Pro the same question and a Claude arbiter flags AGREED vs DISAGREED — the contested stat gets marked, not buried.
Cardless users
Want cited search and a second opinion without a foreign card
Both run inside Telegram — no card of any country involved. Search for sources, Council for cross-model consensus.
How it works01–03

How the source check and cross-check work today

  1. 01

    Ask @vustSearchBot for a cited answer

    Send your question and pick Research mode. The answer comes back with inline [N] markers next to specific sentences and a Sources button opening the matching links. Quick and Standard are the fast lookups: no inline [N] markers, but the source list still opens behind the Sources button.

  2. 02

    Open the sources behind a claim

    Each [N] maps to a link you can click and read — real per-claim provenance, so you verify a sentence instead of trusting a source-free paragraph.

  3. 03

    Get a second opinion from the Council

    For a contested claim, run it through the Council inside @vustbot: three independent models answer the same question and a Claude arbiter returns AGREED and DISAGREED sections, flagging exactly where they diverge.

Same tool · in Telegram@vustSearchBot

Cited search + a 3-model second opinion

Open @vustSearchBot for answers with openable [N] sources; for a cross-check on a contested claim, use the Council inside @vustbot.

Open in Telegram
Quality & trust

Two surfaces, kept apart

Per-claim sources — that's Search, on Research

Research mode gives inline [N] markers tied to statements plus a Sources button of openable links. That is where clickable evidence comes from; sources cannot be added to a Quick answer afterwards.

Answer-level agree/disagree — that's the Council

The Council compares three whole answers (Claude, Grok, Gemini) and flags divergence. It's answer-level consensus, not per-claim provenance — and it does NOT attach web source links. We won't blur the two.

Use the right tool for the job

Need clickable evidence behind a sentence? Search. Want to know whether independent models actually agree before you rely on a number? Council. They're complementary, not the same feature.

FAQ

Frequently asked questions

Is there a 'compare sources' mode I switch on?

No, and we won't invent one. This page describes a task, not a button. What exists is the Research mode, which splits your question into several sub-queries and synthesizes them into one answer carrying inline [N] markers plus a Sources button that opens the matching links. That is where cross-source work actually happens.

Do all tiers show sources?

All three modes can. Whenever an answer has web results behind it a Sources button appears underneath and opens the list — every mode has it. What Research adds is the inline [N] markers that tie a specific sentence to a specific source; Quick and Standard skip those for speed, and they cannot be added to a Quick answer after the fact. Pick the mode by whether you need per-claim provenance, not by whether you want sources at all.

What is the Second Opinion / Council, and how is it different from Search?

The Council of Sages lives inside @vustbot. It sends the same question to three independent frontier models — Claude, Grok (xAI) and Gemini — then an arbiter writes fixed sections including AGREED and DISAGREED. It compares whole answers to flag where the models diverge. It does not attach web source links to each claim — that's what Search is for. Use Search for openable sources; use the Council to see whether independent models actually agree.

Does the Council cite web sources for its claims?

No, and we won't imply it does. The Council's job is answer-level agreement and disagreement across three models — it flags where they diverge, it doesn't pull a numbered list of openable web links. If you need clickable sources behind a statement, run the question through @vustSearchBot in Research mode instead.

Why run both instead of just trusting one answer?

A single model can be confidently wrong. Openable sources let you check the evidence behind a specific sentence; a three-model cross-check tells you whether that sentence is contested in the first place. Together they turn 'sounds right' into 'I can see the source and I know the models agree' — or 'the models disagree here, so dig deeper before I rely on it.'

Do I need a card or an account?

No. Both run inside Telegram, no foreign card needed: your first Research run is free, then you keep going with a VUST plan. Open @vustSearchBot for search; the Second Opinion / Council is inside @vustbot.

Ready when you are

Trust the answer you can trace.

Get openable, cited sources from @vustSearchBot — then flag where three independent models disagree with the Second Opinion Council in @vustbot. Cardless, in Telegram.