Alternative to Consensus
Consensus vs. Agent Bayes: Which One Fits Your Research Needs?
Consensus tells you whether a field agrees. Agent Bayes tells you what your own library supports, with the page behind every claim. Compared honestly.
Choose Consensus if
- You have a yes-or-no empirical question and want to know where the field stands.
- You do not have a library yet and need to search 220 million papers.
- A meter showing how the literature splits is the answer you want.
- You want a cited report generated from a broad sweep in one run.
- You want a cheap paid tier at 15 USD a month.
Choose Agent Bayes if
- You already chose your sources and want answers only from them.
- Claims need author, year, and printed page range for a bibliography.
- You want the argument behind a disagreement, not a count of which side has more papers.
- The work runs long and needs version history, snapshots, and per-claim provenance.
- Your library includes papers in languages other than English.
- You want to check whether your own wording actually matches your cited sources.
Consensus and Agent Bayes at a glance
| Axis | Consensus | Agent Bayes |
|---|---|---|
| What it reads | An index of over 220 million papers, with full text included where available and institutional access connectable. | Only your knowledge bases. Every retrieval stays inside the corpus you curated. |
| Who chose the evidence | A ranker, from the whole index. | You, in advance, by deciding what went into the knowledge base. |
| Output shape | A results page with the Consensus Meter, plus Pro Analysis and Deep Search reports. | A mindmap of cited claims you and the agent develop, plus prose synthesis on request. |
| Citation quality | Each result links to the paper with an extracted finding line. | Author, year, printed page range, and the exact source chunk, opened in a built-in PDF reader at that page. |
| When sources disagree | Counted and displayed as a yes, no, and possibly split across the index. | Preserved as sibling nodes carrying the competing arguments and their evidence, inside your map. |
| Does the work accumulate | Each question produces its own result. Reports are generated per run. | Projects hold multiple mindmaps with conversation history, a rolling version stack, and named snapshots. |
| Depth per source | An extracted finding line, with full text used in analysis where available. | Every page indexed with layout-preserving OCR, semantic chunking, and structured bullet points. |
| Non-English sources | Index coverage is largely English-language. | Bullet points are always written in English regardless of source language. |
| Pricing | Free tier, Pro at 15 USD per month, Deep at 65 USD per month, as of 3 August 2026. | Metered credits charged in 0.25 increments against an inspectable ledger. |
What Consensus gets right
The Consensus Meter is the best single idea in this category. Ask an empirical question and it shows how the literature splits across yes, no, and possibly, drawn from a corpus of more than 220 million papers. For a researcher who needs to know quickly whether a question is settled or contested, nothing else answers it in one screen, and Agent Bayes has no equivalent.
The product has also moved past the abstract-only stage that limited it early on. As of 3 August 2026 full text is used in analysis where it is available, and users can connect institutional access so library subscriptions extend the coverage. Pro Analysis synthesises across papers, and Deep Search runs a broad review over more than a thousand papers and produces a cited report.
Pricing is reasonable for what it does. Free, then Pro at 15 USD a month and Deep at 65 USD, well under what comparable research tools charge.
Where researchers hit the limit
A ranker chose your evidence. This is the central issue with every open-index tool and it is not a bug in Consensus. You asked a question, the system searched 220 million papers, and eight came back. You did not decide the inclusion criteria. You did not check the journals. For a quick orientation that is the right trade. For a chapter you will defend, an evidence base you did not curate is a liability, because the first question is always "why these papers".
A finding line is not a passage. Each result carries an extracted one-line summary of what the paper found. That is useful for scanning and it is not what you cite. Between the finding line and the sentence in your thesis there is a step where you have to open the paper and check, and the tool does not shorten it.
A meter counts, it does not explain. Knowing that fourteen papers say yes and six say no tells you the field leans one way. It does not tell you why the six disagree, which is usually the interesting part and often the reason the six exist. Methodological disputes do not show up as a tally.
Nothing persists. Each question produces its own result page. Ask forty good questions over a term and you have forty result pages and no place where your understanding is being built.
How Agent Bayes approaches it differently
You set the boundary. Knowledge bases are libraries you assemble. Every retrieval stays inside them, and you can scope any query further by tag, author, year, or publication. A single knowledge base serves several projects and a project can draw on several at once.
Depth on every page, not a summary line. Each PDF is run through OCR that preserves multicolumn text, tables, figure captions, footnotes, and headers, then split into semantic chunks, then distilled into structured bullet points that carry enough context to stand alone. Those bullet points are always written in English regardless of the source language, so a German methods paper is searchable next to an English one. A limitation buried in an appendix is as retrievable as the abstract.
Disagreement as structure. When your sources conflict, the competing positions become sibling nodes on the map, each with its own evidence chain, positioned next to each other rather than resolved. Ask for a prose discussion and the agent produces a cited passage that moves through each school of thought and shows where the evidence actually diverges. That is the shape of a literature review, and a meter cannot produce it.
Evidence you can hand to a reviewer. Every claim links to author, year, printed page range, and the exact chunk of source text, and clicking opens the PDF at that page. Complex syntheses carry several citations on one claim.
Work that compounds. Projects hold several mindmaps, each with its own conversation history, a rolling version stack for undo and redo, named snapshots that survive outside that stack, and provenance on every change.
Checking yourself. Select any nodes and Agent Bayes reads the full text behind each attached citation, scores how well it supports the claim as written, and tightens phrasing that overstates.
Alternative to Consensus, or the step after it?
Yes, and the sequence is natural. Consensus is where you find out whether a question is contested and which papers are doing the arguing. Agent Bayes is where those papers become an argument with every claim traceable to a page.
What Agent Bayes will not do: search the open literature, tell you how a whole field splits, or work without a library. If you have no papers yet, start with Consensus.
Questions researchers ask about Consensus
Does Agent Bayes have anything like the Consensus Meter?
No. The meter measures agreement across a whole index, which requires reading the whole index. Agent Bayes reads only your library, so a meter over it would only tell you about the papers you already chose. What Agent Bayes does instead is show you the shape of the disagreement inside your corpus, as sibling nodes carrying each position and its evidence.
Is a 220 million paper index not strictly better than my own library?
For finding out what exists, yes. For work you have to defend, the size is also the problem, because you did not vet what came back. A ranker decided which eight papers answered your question. In a curated knowledge base you set the inclusion criteria before anything was retrieved, and you can narrow further by tag, author, year, or journal.
Consensus now reads full text. Does that close the gap?
It narrows it. Reading full text where available is a real improvement over abstract-only extraction. The remaining differences are provenance grade, printed page ranges versus paper-level links, and persistence, because a Consensus result is generated per question and does not become a workspace.
Which is better for a literature review chapter?
Use Consensus early to find out whether a question is settled and to locate the papers that matter. Use Agent Bayes to build the chapter, because a review needs a structure with traceable claims, not a series of search results.
Can I use both?
Yes. They sit at different points in the same workflow. Consensus orients you across the field. Agent Bayes turns the papers you kept into an argument you can defend.
Bring your own papers and see what they actually say
Index your library, ask the agent to map it, and open any claim at the page it came from. Zotero users can send papers across without re-uploading anything.