Loading the registry…
Start here
Pick what you are trying to do. Each theme recommends two or three tools that work well across the islands, with the rest one click away. Everything in the registry is under All tools.
What do you need to do?
Who offers what
Every tool in the catalogue against every institution. Most cells are empty because most questions have not been asked yet; each IT conversation fills a column.
Institution profiles
The questions for a meeting with an institution's IT or library staff, and the answers so far. UA answers are pre-filled drafts until UA confirms them; entries marked "checked" come from public registries.
AI and your data
Everything typed or uploaded into a cloud AI tool leaves your computer and is processed on the vendor's servers. That holds for chat assistants and for coding agents such as Claude Code, which run on your own laptop but send your prompts and every file they read to the vendor. So the useful question is what happens to the data once it arrives: is it used to train models, how long is it kept, and does an agreement with your institution cover it?
Three ways to use AI, by where the data goes
Stays on your computer
Open models run on your own machine with LM Studio or Ollama, and Whisper-based tools such as noScribe transcribe interviews offline. Nothing leaves the device. The models are weaker than the hosted ones and need a reasonably recent laptop, about 16 GB of memory.
For confidential material: interview transcripts, student records, health data.
Leaves, under your institution's agreement
Microsoft Copilot Chat signed in with a work or school account, Gemini under Google Workspace for Education, or an institution-wide ChatGPT Edu or Claude for Education account. Data is processed by the vendor, but under a contract with the institution: not used to train models, kept according to the agreement, and the institution controls the accounts.
For internal work: drafts, internal documents, published data, teaching material. Personal data only where the institution's policy allows it and consent covers the use.
Leaves, under your own personal terms
A free or paid personal ChatGPT, Claude or Gemini account. The terms are between you and the vendor. ChatGPT's personal plans train on conversations unless you switch it off; Claude's personal plans ask you to choose, and keep conversations for five years if training is on and 30 days if it is off. The institution has no say in any of it.
For public information only.
Which data, which tool
| The data | Examples | Where it may go |
|---|---|---|
| Public | Published reports, open statistics, your own published articles | Any of the three |
| Internal | Drafts, minutes, internal documents, course material | Your own computer, or an institutional account |
| Personal or confidential | Student records and grades, interview transcripts, health or HR data, unpublished data about people | Your own computer. An institutional account only if the policy explicitly allows it and the consent people gave covers it |
If your institution has no AI accounts yet
Check what the existing licence already includes before buying anything. Microsoft 365 education plans (A1, A3 and A5) include Copilot Chat at no extra licence cost for staff, and for students aged 13 and over once an administrator switches it on. Google Workspace for Education includes the Gemini app and Gemini Notebook in every edition. Switching these on gives people a protected option. Without one, staff and students keep using personal accounts, and institutional data goes with them.
Two settings are worth asking IT about. Copilot Chat sends its web searches to Bing under consumer terms, and an administrator can switch that off. And outside the EU, Microsoft may route the processing to other regions when capacity is short.
AI in teaching and assessment
Detectors that claim to spot AI-written text are unreliable as evidence. A study in Patterns (Liang et al., 2023) found that detectors flagged 61 percent of essays by non-native English writers as AI-written, against very few essays by native writers. OpenAI withdrew its own detector in July 2023 for low accuracy. In multilingual Caribbean classrooms, those false positives land on the students least able to contest them. A detector score can open a conversation with a student and should never decide a case on its own. Assessment design, such as drafts handed in along the way and a short oral follow-up, does more.
A policy can start small
Four decisions cover most of it: which accounts are sanctioned, which data may go into which tool (the table above), how AI use is acknowledged in student and staff work, and how suspected misuse is handled without relying on detectors. UNESCO's guidance on generative AI in education and research is a common starting point. A September 2026 UNESCO IESALC survey of 200 higher education institutions in Latin America and the Caribbean found that 87 percent use AI and 26 percent have a formal strategy, so an institution without one is in the majority.
The tools themselves are in the catalogue under Use AI tools safely and Set AI rules for teaching.
Sources, checked 3 October 2026
- Microsoft: Copilot Chat privacy and protections; availability and eligibility
- Google: Gemini and Gemini Notebook for Education
- OpenAI: data controls in ChatGPT; ChatGPT Edu; classifier withdrawal
- Anthropic: consumer terms; commercial data retention; Claude Code data usage
- Local tools: LM Studio offline operation; Ollama cloud models
- Liang, W., Yuksekgonul, M., Mao, Y., Wu, E. and Zou, J. (2023). GPT detectors are biased against non-native English writers. Patterns. doi:10.1016/j.patter.2023.100779
- UNESCO IESALC (2026). AI implementation in higher education in Latin America and the Caribbean
Vendor terms change often, and two products were renamed in 2026 (Microsoft 365 Copilot Chat became Microsoft Copilot Chat; NotebookLM became Gemini Notebook). Check the source before relying on a detail.
About this registry
It started with a question from IPA: which platforms work well for storing, publishing and analysing data? The answers existed, spread over four institutions and a handful of people. The registry puts them in one place, as the institutional half of the DCDC baseline assessment. The survey measures what individuals need; the registry records what institutions offer, to whom, and with what support.
Reading the institution badges
A dashed badge means nobody has asked that institution yet, which is different from "Not offered". A badge marked draft has not been confirmed by the institution.
FAIR ratings
The four letters rate the tool, never the institution: how well does it help people produce outputs that are Findable, Accessible, Interoperable and Reusable?
- F
- Findable: persistent identifiers and rich, indexed metadata.
- A
- Accessible: retrievable through open standards without the tool's own licence, with metadata that outlives the data.
- I
- Interoperable: open file formats and shared standards such as CSV, XLSForm and REFI-QDA.
- R
- Reusable: licences, provenance, and scripted work that can be repeated.
Strong does this well by default Partial depends on how it is used Weak works against it n/a not rated, because it is not what the tool is for
Findable and Accessible are rated only for tools that publish, host or identify research outputs: repositories, identifier services and code platforms. A file becomes findable through the repository it is deposited in, with a DOI and indexed metadata, not through the storage, analysis or dashboard software that produced it. So Microsoft 365, SPSS or Power BI show a grey n/a there. That is not a shortcoming. It marks the point where the work has to move to a repository. Working storage gets n/a for Reusable too, since the licence and provenance are set at deposit.
The ratings were last reviewed on 7 October 2026. They are the network's working judgement; if you use a tool and disagree, tell us.
Ground rules
The registry describes and does not rank. No institution gets a score, for the same reason the FIMM design uses profiles instead of an index: in a network this small, a league table costs the goodwill of the partners it measures. Each institution's column is filled from that institution. Tool assessments are the network's working judgement and open to correction by anyone who uses the tool. Costs appear as a model instead of an amount, because prices vary by region and change every year.
Where the data comes from
This page shows a dated snapshot of the registry; the bar at the top gives the version and date. Corrections and additions go to the network administrator and appear in the next snapshot.