> Import references. Deduplicate. Screen abstracts.
> Process PDFs. Extract structured data.
> Native providers + OpenAI-compatible profiles.
> Workspace mode persists AI suggestions and human decisions.
> Origin-aware counts distinguish imports from PDF-only records.
> Desktop GUI & browser Web App included.
> Config-aware cache + JSONL audit ledger.
> One local-first workflow. Human verification stays in control.█
SYS: Windows // macOS // Linux • REQ: Python 3.10+ • LIC: MIT
SLR Assistant is an open-source, local-first tool that helps researchers conduct systematic and scoping reviews with AI-assisted screening and extraction.
Assists repetitive review tasks: importing references, identifying likely duplicates, screening titles/abstracts, processing PDFs, recording human decisions in a local workspace, extracting structured fields, and exporting auditable outputs.
PhD students, researchers, professors, evidence synthesis teams, and authors of systematic or scoping reviews who want a local workflow with transparent AI assistance.
Not a replacement for human reviewers, not a public cloud service, not a multi-user SaaS platform, not a guarantee of PRISMA compliance, and not a clinical or policy decision system.
// Supports PRISMA-aligned workflows from import to extraction table. Human verification remains required.
Import RIS, BibTeX, or CSV exports from PubMed, Scopus, Web of Science. Auto field normalisation.
Helps identify likely duplicates using DOI and fuzzy title matching, with final review under researcher control.
AI-assisted title and abstract screening before PDF download, producing decisions and rationale for human review.
Local project folders persist imports, PDF metadata, records, review items, decisions, and audit events in SQLite, with clear Workspace Mode versus Legacy Mode labels.
AI output is stored as a suggestion; final Include / Exclude / Maybe status is saved as a human decision. Queue filters show visible versus total review items.
Uses a multi-engine PDF extraction cascade for many common PDF formats, with extraction status recorded.
Quote-Then-Answer prompts encourage evidence-grounded outputs; human verification is still required.
Stage 1: fast include/exclude. Stage 2: detailed criteria. Supports PRISMA-aligned workflows.
Multiple PDFs can be processed with configurable workers and provider-aware rate-limit settings.
Export structured tables for review, verification, workspace reporting data, PRISMA-ready counts, and methods disclosure.
Workspace exports include decision CSV/XLSX files, AI and human decision logs, full-text exclusion reasons, PRISMA-ready counts, and a manifest.
Your criteria, your extraction fields, or a starting point from built-in domain templates.
Browser-based local interface with Workspace create/open/close, PDF upload, progress counts, review queue, workspace exports, and local-only server binding.
Optional AI assistance can help draft screening criteria, prompts, and extraction fields using your selected provider.
Configuration-aware JSON cache and JSONL audit ledger record provider, model, prompt hash, text hash, cache key, and cache hit/miss. Audit logs do not store API keys or full paper text.
// Choose local models, direct cloud providers, router profiles, or a custom OpenAI-compatible endpoint.
| PROVIDER / PROFILE | PRIVACY MODE | MODEL SELECTION | BEST FOR |
|---|---|---|---|
| Ollama | Local only | Local model ID from your Ollama installation | Most privacy-preserving path; offline/local workflows |
| OpenAI | Direct cloud | Recommended presets plus manual model ID | Direct provider workflow with current OpenAI models |
| Anthropic Claude | Direct cloud | Recommended presets plus manual model ID | Claude workflows where reasoning style fits the protocol |
| Google Gemini | Direct cloud | Recommended presets plus manual model ID | Google Gemini users; prefer stable model IDs for reproducibility |
| DeepSeek / Mistral / Kimi / xAI | Direct cloud | Profile defaults plus manual model ID | OpenAI-compatible provider profiles without one-off classes |
| OpenRouter | Router / third party | Router model slug or alias | Comparing many providers through one router account |
| LM Studio / vLLM / LocalAI | Local only | Local OpenAI-compatible model ID | Self-hosted local or lab-managed inference endpoints |
| Custom OpenAI-Compatible | Custom endpoint | Any compatible model ID supported by your endpoint | Institutional, private, or experimental compatible APIs |
> For maximum privacy, use Ollama or another local endpoint you control. Router providers such as OpenRouter send paper text through an additional third party before reaching the selected model provider. Built-in model lists are recommendations only; manual model IDs remain supported.
// 4 steps from raw references to review-ready data for human verification
Load RIS/BibTeX/CSV exports. Auto-normalise fields. Identify likely duplicates using DOI and fuzzy title matching.
AI evaluates each title and abstract against your criteria and produces Include / Exclude / Maybe recommendations. No PDFs needed yet.
Drop PDFs into a folder. Text extraction, two-stage screening, configurable parallel processing, and configuration-aware JSON caching.
AI drafts structured extraction fields into Excel/CSV exports for review, verification, and reporting.
// Pre-configured criteria for common research domains
Any research field
RCTs, PICO, cohorts
Pedagogy, outcomes
Climate, ecology
Algorithms, benchmarks
// Two interfaces — Desktop GUI & browser Web App. Same setup, your choice.
REM 1. Download or clone the repository
git clone https://github.com/sadeghanisi/SLR.git
cd SLR
REM 2. One-click setup (creates venv + installs all deps)
setup.bat
REM 3a. Launch desktop GUI
launch_gui.bat
REM 3b. OR launch Web App (opens in your browser)
cd WebApp
run.bat
REM → then open http://127.0.0.1:5000# 1. Download or clone the repository
git clone https://github.com/sadeghanisi/SLR.git
cd SLR
# 2. Create venv & install
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
# 3a. Launch desktop GUI
python slr_gui.py
# 3b. OR launch Web App (opens in your browser)
cd WebApp
python app.py
# → then open http://127.0.0.1:5000# Requires Python 3.10+
python -m venv .venv
# Activate
.venv\Scripts\activate # Windows
source .venv/bin/activate # macOS / Linux
# Install all dependencies
pip install -r requirements.txt
# Option A — Desktop GUI
python slr_gui.py
# Option B — Web App (runs locally in your browser)
cd WebApp
python app.py
# → open http://127.0.0.1:5000# Step 1 — Clone and set up (same as desktop GUI)
git clone https://github.com/sadeghanisi/SLR.git
cd SLR
python -m venv .venv
# Activate venv
.venv\Scripts\activate # Windows
source .venv/bin/activate # macOS / Linux
# Step 2 — Install all dependencies (Flask is included)
pip install -r requirements.txt
pip install flask
# Step 3 — Launch the Web App server
cd WebApp
python app.py
# Step 4 — Open in browser
# → http://127.0.0.1:5000
# Windows shortcut: double-click WebApp\run.bat
╔════════════════════════════════════════════════════════════╗
║ BUILT FOR REAL RESEARCH ║
╠════════════════════════════════════════════════════════════╣
║ ║
║ PhD students processing 500 papers for a dissertation ║
║ Professors supervising multiple systematic reviews ║
║ Research teams running large-scale evidence synthesis ║
║ ║
║ >> AI assists repetitive tasks; reviewers decide. << ║
║ ║
╚════════════════════════════════════════════════════════════╝
The software is free and open source under the MIT License. Cloud AI providers may charge for API usage. Ollama/local models can run without provider API fees, depending on your own hardware.
NO. Graphical interface — click buttons, select options, browse files. Zero command-line knowledge needed after initial setup.
LLM = Large Language Model (the AI that reads papers, like ChatGPT). API key = password letting the tool talk to the AI service. Sign up with any provider to get one. See the COMPLETE_USER_GUIDE.md.
For maximum privacy, start with Ollama or another local endpoint. For cloud workflows, choose a direct provider whose privacy and pricing terms fit your project. Router providers such as OpenRouter add an additional third party in the data path.
Accuracy varies by topic, criteria clarity, model, prompt, PDF quality, and corpus. Quote-Then-Answer prompts encourage evidence-grounded outputs, but human verification is required.
The queue may be filtered by stage, status, or record origin. For example, 900 imported references can show only 5 visible rows when Status is Suggested or Origin is PDF-only. Clear filters to show all review items.
Yes, as an AI-assisted workflow. Workspace exports can generate PRISMA-ready counts and methods disclosure text, but you remain responsible for checking counts, final inclusion/exclusion decisions, extracted-data verification, and reporting.
INPUT: RIS, BibTeX (.bib), CSV, PDF • OUTPUT: Excel (.xlsx), CSV, JSON, Markdown methods disclosure, plain-text summary reports, and workspace export manifests.
Cloud provider: submitted text is sent to that provider's API. Router provider: submitted text also passes through the router. Ollama/local endpoint: paper text can remain on your computer or local network, depending on how you host it.
> "As Is" Provision: This software is provided "as is," without warranty of any kind, express or implied. The authors and distributors accept no responsibility for decisions made based on AI-generated screening or extraction results.
> AI Limitations: This tool assists with — but does not replace — human judgment. AI models can and do make errors, including incorrect inclusion/exclusion decisions and inaccurate data extraction. All AI outputs must be independently verified by qualified researchers before use in any publication, thesis, clinical decision, or policy document.
> No Academic Guarantee: Use of this tool does not ensure compliance with PRISMA, CONSORT, or any other reporting standard. Researchers remain solely responsible for the methodological integrity, transparency, and accuracy of their systematic reviews.
> Data Privacy: When using cloud-based AI providers (OpenAI, Anthropic, Google, DeepSeek, OpenRouter, or others), your paper content is transmitted to third-party servers. Router providers such as OpenRouter add another third party. The authors of this tool make no representations regarding how those providers store, process, or use your data. Consult each provider’s privacy policy before processing sensitive or unpublished material. For confidential data, use the local Ollama option.
> Cost and Billing: API usage fees are charged directly by third-party AI providers. The authors of this tool have no visibility into, or responsibility for, charges incurred through your API account. Monitor your usage and set billing limits with your provider before running large processing jobs.
> Institutional Compliance: It is your responsibility to verify that AI-assisted research methods comply with your institution’s policies, your funding body’s requirements, and the ethical standards of your field. Disclose AI tool usage in all relevant sections of your research output.
> No Liability: To the fullest extent permitted by law, the authors, contributors, and distributors of this tool shall not be liable for any direct, indirect, incidental, or consequential damages arising from its use, including but not limited to data loss, incorrect research conclusions, academic penalties, or financial charges.
By using this tool, you acknowledge that you have read, understood, and accepted these terms.
┌─────────────────────────────────────────────┐
│ READY FOR AI-ASSISTED REVIEW? │
│ Download → Set up → Review with care. │
└─────────────────────────────────────────────┘
v3.5.2 • MIT License • Python 3.10+ • WIN / MAC / LINUX