Features
Every feature built around researcher needs
Not a general-purpose AI wrapped in a research label. Each capability addresses a specific failure mode researchers face — from the first upload to the final citation check before submission.
Your evolving research workspace
The real problem
Most research tools deliver a finished output and leave. Real research never stops at a single synthesis — you return, revise, deepen, and challenge your own structure as understanding grows. A static document cannot keep up with an evolving argument.
Full history, always.
Every version of your mindmap is preserved with undo, redo, and complete provenance for every change — whether made by the agent or by you.
- Direct the agent to expand any branch, investigate a new angle, or strengthen claims with more evidence at any time.
- Edit any node's text, rearrange the structure, assign or remove citations, and embed images alongside text in any node.
- Each project holds multiple mindmaps organized around different research questions within the same workspace.
- Custom instructions teach the agent domain-specific terminology, methodological distinctions, and theoretical priorities before each session.
- Return days or weeks later and pick up exactly where you left off — full conversation history and version stack intact.
From question to synthesis in minutes
The real problem
Research synthesis is not just retrieval. It is the cognitively demanding work of integrating, comparing, and structuring findings from dozens of sources. General-purpose chatbots do not hold your citation chain or maintain structured outputs across an entire research conversation.
Up to 3 self-correction passes.
The pipeline evaluates its own output for gaps and reruns before presenting a result. You receive synthesis that has already been reviewed, not a first draft.
- Natural language instructions — tell Agent Bayes what to research, not how to search.
- Four specialized agents coordinate: FrontDesk interprets intent, Researcher retrieves and ranks evidence, Editor structures the mindmap, and RePlanner self-corrects gaps.
- Research depth controls how broadly the agent searches, from a focused pass to a multi-pass sweep.
- Choose the language model and reasoning depth before each run to balance speed, quality, and credit cost.
- Contradicting viewpoints across sources appear as distinct nodes, never collapsed into false consensus.
Every claim traced to its exact origin
The real problem
The core credibility risk in AI-assisted research is claims that sound plausible but cannot be sourced. Reviewers and supervisors will notice. Your argument structure depends on provenance that can be traced all the way back to the passage.
No exceptions.
Every AI-generated node links to the specific paper, author, year, page range, and exact text chunk it drew on. Click any citation to read the evidence directly.
- Citations include author, year, page range, and the exact chunk of text the agent drew on.
- Click any citation to open the source passage in the PDF reader without leaving your mindmap.
- AI citation verification reads the full text behind each assigned citation and scores how well the evidence supports the claim.
- Where phrasing overstates or misrepresents the source, the agent rewrites the node to bring it into tighter alignment while preserving your intended meaning.
- Scholarly disagreements appear as distinct nodes with independent evidence chains — never smoothed into false consensus.
A knowledge base built for academic content
The real problem
Academic PDFs are not plain text. Multi-column layouts, footnotes, tables, and figures are routinely corrupted by naive extraction. A corpus mixed across research questions produces noisy, poorly scoped retrieval that misleads rather than informs.
Any language. No barriers.
Papers in French, German, Spanish, Japanese, or any other language are indexed in English and become fully searchable without any manual translation step.
- Each page undergoes careful OCR that preserves the original layout: multicolumn text, tables, figure captions, footnotes, and headers.
- Documents are segmented into semantic chunks, each processed into structured English bullet points that stand on their own as searchable units.
- Knowledge Bases are project-scoped: attach multiple bases to a project, or share one base across several projects simultaneously.
- Document tags act as precision retrieval filters, scoping any search or agent query to only the papers relevant to a specific question.
- Add, update, organize by topic, or remove documents at any time — the corpus is always under your control.
Search, read, and collect evidence in one place
The real problem
Finding the right passage, reading the source, and filing the evidence are three separate workflows in most research setups. Friction between them means insights get lost and annotations go unused when it matters most.
One interface. No tab switching.
Semantic search, PDF reading, annotation, citation assignment, and thematic labeling all live in the same place — accessible without leaving your mindmap.
- Semantic RAG search queries your entire corpus in natural language, filtered by tags, author, or publication year.
- Add any search result directly as a citation to a mindmap node, or label it for thematic grouping across the project.
- The built-in PDF reader shows AI-extracted bullet points in a synchronized sidebar that keeps pace as you scroll.
- Highlight passages with color coding, attach icons to categorize findings by type, and add free-text annotations to capture your thinking in context.
- Labels let you curate named collections of evidence across your entire project for cross-document thematic analysis with multi-label filtering.
Ready to try it
See it in your own research
Join researchers who are already building better literature reviews with Agent Bayes.