Documentation
CciipScope
An open-source workbench for research-grounded paper understanding, conference-aware review, and actionable manuscript revision.
Install and run →
Prepare Docker, configure a model, and create the first paper workspace.
Understand the workflow →
Follow the evidence flow through every analysis stage and output panel.
Projects and data →
Review input formats, access tokens, retention, and how project data is used.
Configure the system →
Models, MinerU, search, retention, compilation, and project storage.
What CciipScope does
A paper project begins with a PDF or LaTeX archive and a target conference. CciipScope parses the manuscript into a structured internal representation, discovers relevant literature, runs mechanical and scientific analyses, and presents all results in a private token-scoped workspace.
The workflow is deliberately fixed rather than conversational. This makes the analysis surface predictable: users can inspect the paper structure, research landscape, compliance findings, method analysis, macro logic, full review, section feedback, and revision checklist without following an agent's internal execution trace.
Design principles
Evidence before judgment
Novelty and positioning claims should be tied to retrieved literature. The system distinguishes source evidence, model inference, and unresolved uncertainty.
Multiple views before synthesis
Specialized review passes examine background and originality, methodology, logic and consistency, and results and writing. A final deliberation integrates these views instead of concatenating them.
Revision over verdict
The score and recommendation are only one output. Method alternatives, cross-section conflicts, local section feedback, and a prioritized checklist are designed to make the result useful after the review.
Private by project
Each workspace uses a high-entropy bearer token. Project artifacts and LLM interaction logs stay in the configured data directory, and inactive projects are removed after the retention period.
Current scope
- English-language AI and computer-science manuscripts.
- Conference policies for supported AI and NLP venues.
- PDF and LaTeX ZIP input, with explicit main-file selection for LaTeX projects.
- Model providers exposed through Anthropic-compatible or OpenAI-compatible APIs.
- Self-hosted SearXNG and Dockerized MinerU integration.
Choose a path
For a first local run, continue to Getting started. Contributors should also read the repository's architecture and contribution guides after completing one analysis end to end.