Local life
Public services, transport, planning, heritage or local history
A practical project for curious people
Build a knowledge base your AI can trust
Choose any subject. Turn reliable sources into a small, linked knowledge bundle. Test what your AI knows. Then build an app that helps somebody else learn.
- No prior coding or linked-data knowledge assumed
- Free, static and usable with your choice of AI
- Privacy and evidence decisions before upload
OKFOpen Knowledge Format: small Markdown files with machine-readable facts.
Semantic linkA named, directed connection with evidence — not just a clickable link.
GroundingGiving an AI the relevant evidence and checking that its answer stays inside it.
MCPModel Context Protocol: an optional way for an AI to request bounded context.
Learn by making
Eight small stages, one useful result
Each stage repeats the same rhythm: explain, inspect a worked example, do one task, check it, then recall what mattered. You can stop after any checkpoint and return later.
- 01
Choose
Pick a subject small enough to finish and name who it should help.
- 02
Question
Write the questions your bundle and app must answer before collecting data.
- 03
Research
Find sources and decide authority, rights, privacy, freshness and gaps.
- 04
Model
Give things stable identities and add only relationships you can explain and evidence.
- 05
Build
Author a small OKF 0.2 bundle, then use exact checks instead of guesswork.
- 06
Explore
Use Reader, Search, Links, Graph, Timeline and Inspect to find defects.
- 07
Ground
Connect your AI and compare its answers with questions you held back.
- 08
Create
Predict, run, inspect and modify working code before making your own learning UI.
Your subject, not ours
Start with something you care about
A useful first bundle is narrow enough to review yourself: roughly 15–60 concepts and 5–15 questions.
Culture
Music, films, books, games, sport, art or a personal collection
Study
A course topic, scientific field, historical period or reading list
Work
Policies, APIs, guidance, research papers or organisational knowledge
Private
Your own notes or records — kept local unless you have a safe sharing decision
Working examples
Learn from real OKF bundles
These entries come from the governed registry. Their labels such as preview, candidate or bounded demonstrator are part of the evidence, not decoration.
AI Infrastructure
Default reference bundle generated from the AI infrastructure Markdown corpus during migration.
- Version
- v0.4.0
- Delivery
- local-sample
ONS data discovery OKF
Metadata-only ONS discovery demonstrator with explicit coverage, confusable alternatives, standards evidence and MCP selection bindings.
- Version
- v0.2.0
- Delivery
- large-corpus
GOV.UK CKAN large corpus
External large-corpus exemplar using chunked CKAN metadata and static search shards.
- Version
- v1.0.0
- Delivery
- large-corpus
UK Government APIs OKF
Multi-source API and data-access large-corpus exemplar.
- Version
- v0.4.0
- Delivery
- large-corpus
UK Legislation complete work catalogue
Complete legislation.gov.uk work catalogue with live progressive CLML discovery.
- Version
- v0.2.0
- Delivery
- large-corpus
UK Whole-Law OKF
Federated overview of independently governed UK legal-source bundles with explicit authority, coverage, freshness and alternate access routes.
- Version
- v0.3.0
- Delivery
- federation
Coventry and Warwickshire Heritage Evaluation
Independently published Evaluation Foundry exemplar covering the faithful source-backed corpus, tiny assurance subset and isolated synthetic capability supplement; promotion evidence is carried by its signed release envelope.
- Version
- v1.0.0
- Delivery
- large-corpus
See the full LLM-Wiki and OKF lineage, including products that are not in the current registry.
Meet the primary learner
Sam is 18, curious and new to this
Sam can browse, edit files and ask an AI for help. They do not yet know data modelling, provenance or deployment. They need a visible result early, plain language, low-cost tools, safe choices and a clear definition of “done”.
“Show me why this idea helps my subject, let me try it on a real example, and give me a check I can trust.”Read Sam’s complete persona and user stories
Your AI is a collaborator, not your evidence
Understand, apply, create — then verify
Ask where an answer came from
Learn identities, sources, dates, rights and uncertainty before asking the AI to make anything.
Compare expected and observed answers
Hold questions back, require record citations and count unsupported or near-neighbour answers.
Read and change working code
Predict, run, investigate and modify a starter before using AI to create your personal interface.