Discuss
Start in chat
Scope the idea, ask for a literature sweep, pressure-test the angle, and decide whether the project is worth writing.
Research writing agent
Cite-Agent
Claude-style research workbench for building a paper, with citations, source trails, empirical outputs, and model decisions kept beside the manuscript.
Chat-first like Claude, but the paper stays visible.
Formatting calm like Google Docs, with real page polish.
Trace rails like VS Code, only when evidence needs review.
One research loop
The chat is where the work starts. The project is where the work becomes durable.
Cite-Agent should not feel like a dashboard you manually operate. Talk to the agent, then inspect the paper, evidence, data, and decisions that the agent leaves behind.
Collect
Keep the research base
Papers, notes, exclusions, and source decisions stay attached to the project instead of disappearing into a chat transcript.
Validate
Run the empirics
Datasets, model outputs, figures, and interpretation notes become part of the same evidence trail as the citations.
Write
Draft the manuscript
The paper is the artifact. The agent drafts, revises, cites, checks, and records how each section was built.
Manuscript evidence
Introduction
4
Sources
linked to claims
2
Claim audits
need support
12
Model edits
diffs recorded
Claim
Readers can inspect which source supported a sentence.
Evidence
Human-AI provenance in scientific authoring, 2023.
Decision
DeepSeek revision pass tightened the claim and added source pressure.
Why it is not just a PDF tool
The output is the paper plus the proof of how the paper was built.
A normal chatbot gives you prose and leaves you to clean it up elsewhere. Cite-Agent keeps the documentation: cited papers, rejected sources, empirical runs, comments, and model edits.
Manuscript surface
Calm like Docs. Agentic like Claude. Traceable like an audit log.
The editor stays simple because the agent does the heavy lifting. Formatting controls remain where they help polish the final paper; provenance and comments live in the rail when you need to inspect them.
Page polishCitation checksClaim supportDecision history
Manuscript
Ask Cite-Agent to revise, cite, or format this section...
Introduction
Literature Review
Methods
Discussion
Introduction
Generative writing systems can draft academic prose, but most tools treat the paper as an export rather than a traceable research artifact [S1].
Cite-Agent keeps evidence, review state, and decision history attached to the manuscript so a reader can inspect why each claim exists.
The result is a paper that can be polished like a document and defended like a research record.
Chat
Insights
Polish phase
Introduction is grounded. Discussion needs one source.
Check the grounds
Format for submission
Find the soft spot
Build the paper and its evidence trail in one place.
Start with a research idea, then let the agent help you search, analyze, draft, cite, check, and polish.