Local AI and Māori GIS

Use AI on your own computer to search documents, find useful kōrero and prepare checked information for mapping without making cloud AI the default destination for your files.

Imagine a folder containing several public Waitangi Tribunal reports. You want to find passages about an area of whenua, identify candidate ingoa wāhi, preserve the wording used in each source, check the places and create a small dataset for QGIS or Google Earth.

Your documents → local document search → local LLM → candidate information → human review → geographic source checking → GIS data → map

The AI helps search, retrieve and structure information. The people doing the mahi decide what the findings mean and how they are used.

If terms such as LLM, inference, embeddings, vector database, RAG, context window, multimodal model or agent are unfamiliar, read AI basics for Māori GIS first. It explains the concepts and how they differ from familiar GIS ideas such as vector geometry, geodatabases and conventional spatial analysis.

Before you load documents

Local processing can reduce the amount of working material sent to external AI providers, but local does not automatically mean sovereign, authorised or appropriate. The important questions still include who has authority over the information, why it is being used, what new information can be inferred, who may see the outputs and whether anything should be mapped or published.

A document being publicly downloadable also does not mean that every piece of kōrero inside it is culturally unrestricted, accurate, current or suitable to republish as a precise spatial feature. Public reports can contain whakapapa-linked information, historical names, evidence about sensitive places and statements made for a particular inquiry or kaupapa. Keep the source and context attached to anything extracted by AI.

Before installing software or loading a document collection, read Before you use AI with Māori information. It brings together Māori-led governance frameworks, organisational approval questions, privacy and public-sector guidance, and prompts for deciding whether the workflow should change.

For Māori-led guidance, start with Te Mana Raraunga, the Te Kāhui Raraunga Māori Data Governance Model, the Māori AI Governance Framework and the Global Indigenous Data Alliance CARE Principles. Then use the practical Māori Data Sovereignty for GIS section on this site for spatial questions about authority, sensitive locations, access, publishing, cloud services, local systems and derived information.

Follow the practical path

1. AI basics for Māori GIS
2. Before you use AI with Māori information
3. Why use local AI?
4. Choose your software
5. Install local AI
6. Search your own documents
7. Working with PDFs
8. Find ingoa wāhi
9. Check the places
10. Turn results into GIS data
11. A complete example
12. Practical AI applications
13. Useful prompts
14. What can go wrong?
15. Downloads and resources

Where this connects to GIS research

For a larger historical workflow, use Archives to map. It combines document search with source registers, place-name checking, historical maps, georeferencing and source-linked GIS features.

The Historical ingoa wāhi worked example shows the complete sequence.

What local means here

For this guide, local means that the language model, document extraction, embedding model, search index and document processing can operate on a computer you control without sending the working material to an external AI provider. AI basics for Māori GIS explains how those components fit together.

That should be visible and testable. You should be able to identify where the model file, documents and index live, whether a cloud provider is configured and whether the workflow still works after the network is disconnected.

AnythingLLM Desktop is the main beginner path in this guide. LM Studio is useful for running and testing local models. Open WebUI is included for people who want a more configurable self-hosted environment. See Choose your software for the current comparison.

What AI is good at

Local models are useful where the evidence is already in your documents but difficult to find manually. They can locate passages across hundreds of pages, compare references, pull repeated attributes into a table and produce a starting list of candidate place names.

They are less useful when source material is missing, scans are unreadable, names are mistranscribed or the answer depends on knowledge the model has not been given.

AI output is a candidate finding. The source remains the evidence.

For more applications including table extraction, plan comparison, draft metadata and QGIS expressions, use Practical AI applications.

From candidate name to map

Preserve the wording from the document first rather than asking a model to supply a coordinate. Then check an identified geographic source such as the New Zealand Gazetteer and any other evidence relevant to the research.

A name not found in the national Gazetteer is not automatically wrong or unreal. Local, historical and customary names can exist outside national datasets. Place names explains how to keep official, historical and locally held names separate.

Keep checking the wider guidance

AI, privacy, public-sector policy and Māori data governance are developing areas. Do not rely on MāoriGIS.nz as the only source. Revisit the Māori-led frameworks and the policies that apply to your organisation or kaupapa, particularly when software, source collections, audiences or intended outputs change.

If newer guidance changes what is appropriate, change the workflow. A different project may require a different model, tighter controls, consultation with relevant people, an approved enterprise service, a non-AI method, or a decision not to map some information.

For wānanga

This section is also designed as the reference path for Ngā Poutama Matawhenua demonstrations. A participant can learn the AI concepts, install one application, add public reports, ask evidence-based questions, identify candidate ingoa wāhi, check selected places, create a GIS-ready table and view the result in QGIS or Google Earth.

See Ngā Poutama Matawhenua resources.

Last reviewed: 4 September 2026