Systematic reviews, from PICO question to forest plot.
Collie brings search, blind dual screening, extraction, risk of bias, meta-analysis in R and PRISMA 2020 onto one desk. AI suggests, with reasons. You decide.
Free in your browser. Nothing to install, no credit card.
Search inside Collie
Strategy translated, results imported from
Statistics in
Nine steps. One trail.
Each step hands off to the next with no spreadsheet exports and no copy-paste, and with an audit trail of who decided what.
Built for the rigor your committee and peer reviewers expect.
Organised around methods described in the Cochrane Handbook, cited as a methods reference. Collie is not a Cochrane product.
PICO search that becomes a boolean string
Build population, intervention and outcome blocks with MeSH terms and synonyms. Collie writes the syntax for each database, applies filters such as the Cochrane HSSS and shows the count before you import.
- Validation against studies you already know should appear
- Eight import formats and deduplication in the same panel
- Forward and backward citation searching through the graph
("Diabetes Mellitus, Type 2"[Mesh] OR "type 2 diabetes"[tiab] OR T2DM[tiab])
AND
("Resistance Training"[Mesh] OR "strength training"[tiab] OR "weight lifting"[tiab])
AND
("Glycated Hemoglobin"[Mesh] OR HbA1c[tiab])
AND
(randomized controlled trial[pt] OR randomi*[tiab])PopulationInterventionOutcomeDesign filter
Meta-analysis in R, without leaving the review
Extracted data feeds the {meta} package directly: binary, continuous, generic, proportion and correlation, with random effects, subgroups, leave-one-out, funnel plots and trim-and-fill.
- A reproducible
analysis.Rto attach to your manuscript - Optional Bayesian sensitivity analysis
- GRADE per outcome with a Summary of Findings table
Network meta-analysis
Compare three or more treatments in one analysis: the network graph, forest plot, league table, inconsistency checks and ranking, computed with {netmeta} from the arms you already extracted.
- The contribution matrix shows how much each study weighs in every estimate
- Forest plot, league table and ranking in one run, with the R script to reproduce it
- Transitivity and certainty stay your judgement: Collie reports the numbers, you decide
- Global and local inconsistency checks (direct vs. indirect evidence) reported next to the estimates
Blind dual screening
Reviewers vote without seeing each other's votes. Conflicts are resolved by consensus or majority, and Cohen's kappa is calculated for you.
κ = 0.82
almost perfect agreement
Risk-of-bias judgements, recorded by domain
Record risk-of-bias judgements by domain, guided by the official tools (RoB 2, ROBINS-I, QUADAS-3 and others), and plot them in robvis style.
Tool names belong to their authors; Collie is not affiliated with or endorsed by them.
A PRISMA 2020 diagram that fills itself in
The flow diagram follows your real decisions, and the 27-item checklist shows what is still missing.
- 1,248 records identified
- 1,016 after removing duplicates
- 87 full texts assessed
- 14 studies included
AI suggests. You decide.
The assisted reviewer reads each record against your criteria and explains its suggestion criterion by criterion. Nothing is saved as a decision until a person accepts it, and the audit trail separates what came from people and what came from the machine.
- Collie
- Justifies with the I1, E2… criteria you wrote
- Generic chat
- Answers from whatever it learned on the internet
- Collie
- Extraction with a quote checked against the PDF
- Generic chat
- Numbers without the page they came from
Assisted reviewer
Suggests include, moderate confidence
- I1 Adults with T2D “patients aged 40 to 65 with type 2 diabetes”
- I2 Supervised resistance training “3 weekly sessions with a physiotherapist”
- I3 HbA1c outcome “change in HbA1c at 12 weeks”
- E1 Non-randomized design Allocation described only in the full text
Take your review to the agent you already use.
Collie's AI runs on the agent you already use, on your own AI subscription. Every plan, Free included, can connect Claude Code, Claude Desktop, Codex, Gemini CLI, Cursor, Perplexity (Mac app) or any other MCP client with the Collie agent kit. Pick yours and copy the setup. The agent works under the same rules as the app: its suggestions stay separate and wait for human approval.
- Plugin with 13 skills, 8 agents, 4 hooks and a live Collie pane
- Connects with a one-time code, no keys to copy
5 databases searched from inside Collie
6 appraisal tools for recording risk-of-bias judgements
27 PRISMA items with automatic status
3 languages: English, Portuguese and Spanish
5 themes, including a color-blind mode
How much time can Collie save?
An honest estimate, not a promise. It depends on your topic, your team and how you use Collie. Below, measured timings are kept apart from estimates built on published studies.
Estimated
18–217 h
estimated time saved on a typical review (3,000 records, 2 reviewers, 30 studies); central estimate about 67 hours.1
Measured
~1.4 s
to import 4,763 records from two RIS files and flag duplicates. All 294 duplicates we planted were flagged; you confirm each group.2
Measured
~6 s
for a random-effects meta-analysis in R with forest, funnel and leave-one-out plots, exported as SVG, PNG, PDF and TIFF.2
- Estimate from published per-task times (e.g. Shemilt 2016, Nussbaumer-Streit 2021, O'Mara-Eves 2015) and our model; conservative to optimistic scenario. Not a guarantee: results depend on topic, team and use of the AI reviewer.
- Measured in October 2026 on one workstation running Collie locally (3 runs, public SYNERGY dataset with planted duplicates; 13 trials from the BCG dataset). The hosted service adds network and queue time.
Start free. Upgrade when the review gets serious.
The Free plan costs nothing and needs no credit card. Every plan can connect your own AI agent via MCP; paid plans add more cloud reviews and more collaborators.
Prices in US dollars. Cancel anytime; your reviews stay exportable.
How AI credits work. Free uses your own agent through MCP, on your own AI subscription. Paid plans add Collie's built-in agent with monthly credits: 1 credit is 1 abstract screening suggestion, and extracting one article costs 20. Run out? Paid plans can buy more inside the app, from 2,500 credits for $4. The agent only suggests; you decide.
Invite a colleague. They get 20% off their first 3 monthly payments, and when they subscribe you get the same on yours. Monthly plans only.
Teach a course or create content? Affiliates earn 30% recurring for 12 months, paid only on revenue received, with their own code. Write to contact@colliesr.com.
Frequently asked questions
Everything you need to know before your first review.
Where is my data stored?
In the cloud, at app.colliesr.com, so your team can work on the same review. When you connect your own AI agent, what the agent reads from Collie goes to the AI provider you chose, under your agreement with that provider.
Can the AI include or exclude studies on its own?
No. Suggestions and extractions are stored separately, with their reasons, until someone on the team accepts them. The audit trail records which decisions were human and which were suggestions.
Do I need to know how to code in R?
No. Collie runs R behind the scenes and shows the forest plot, funnel plot and tables. If you want, download analysis.R and reproduce everything in RStudio.
Can I bring in a review I started in another tool?
Yes. Collie imports RIS, NBIB, BibTeX, CSV and other common export formats, and detects duplicates on the way in.
Is it suitable for a thesis or dissertation?
Yes. The workflow is organised around methods described in the Cochrane Handbook and produces a PRISMA 2020 flow diagram and checklist for your report. Collie is not affiliated with or endorsed by Cochrane or the PRISMA group.
Does Collie do network meta-analysis, and does it rate certainty?
Yes to the network meta-analysis (frequentist, with {netmeta}): estimates, league table, inconsistency checks, ranking and the contribution matrix. Bayesian network models are not available yet. Whether the studies are similar enough to combine, and how certain each comparison is, stays your judgement.
Can I register my protocol on PROSPERO from Collie?
No. Collie drafts the protocol in PROSPERO field order from what your review already has, and you export it and submit it on PROSPERO yourself.
Prefer to talk it through first?
Book a short video call to see Collie on your own question, plan a team or lab rollout, or ask about an institutional plan. Tell us what suits you and we reply by e-mail to confirm a time.
Or write to contact@colliesr.com.
Your next review starts with a question.
Open Collie in your browser, create the review and write your PICO. Collie keeps the trail all the way to PRISMA.
