---
id: ch4-local-kb-poc
type: spine
title: "A local GeoAI knowledge base: the Klatovsko proof of concept"
readingTime: 3
standalone: true
core: false
teaser: "One district, open geodata and community knowledge, readable by people in a book and by AI agents through open files and an MCP endpoint."
voice: universal
parent: null
diagram: diagram-local-kb-stack
recallQ: "What makes the Klatovsko knowledge base grounded, and how can people and AI agents reach it?"
recallA: "Every factual claim is traceable to a named source or data layer, with date, coverage, method and limits stated. People read and co-write it as an interactive book; AI agents read static pages under /kb/, llms.txt, raw markdown and open data files, or query its MCP endpoint at /api/mcp."
highlights:
  - "The unit is a book block tied to a place and to the layers that back it"
  - "People: the interactive book, its author studio and the consortium gallery with editorial lift"
  - "Agents: /kb/ pages, llms.txt, raw markdown, open data files, and an MCP endpoint at /api/mcp"
  - "Success is a small, maintained, checkable base, not a regional oracle"
status: accepted
concept: local-geoai-kb
conceptTitle: "A local GeoAI knowledge base"
parents: geoai|data-to-decisions
lang: en
lens: czech-bavaria
depth: standard
formalism: none
visuality: balanced
lengthBand: deep
genre: explainer
state: edited
carriers: prose|diagram
---

This chapter's last part tests a small idea at full scale: a **local GeoAI knowledge base** for **Klatovsko**, the Klatovy district in the Plzeň Region of Czechia, which is the Czech side of the [Czech–Bavarian pilot](#c/czech-bavarian-pilot). It does not produce another regional profile. It tests whether a book can be the maintained interface between places, evidence and the people deciding about them, and whether AI agents can read the same interface.

## The unit: a block tied to a place

Each block answers one local question about an identifiable geography (municipality, catchment, land block, route or habitat) and names the **open layers** that back it: Copernicus imagery, LPIS, the cadastre, statistics, OpenStreetMap. Blocks state which boundary they use, because datasets often use different ones. The layers themselves are described in [what open data says about Klatovsko](#c/klatovsko-open-data).

## What grounded means

Every factual claim is traceable to a source or layer: publisher, dataset, date, coverage, access link and any transformation. This follows the provenance profile in D4.2, whose priority fields are source, provider, temporal coverage, update status, licence, access level, known limitations and quality status. A mapped overlap is not a cause; satellite greenness is not a diagnosis; local testimony is labelled as testimony. Where evidence is missing, the block says so.

## Two front doors

**For people**, the knowledge base is this **interactive book**: blocks with a recall question, facets to read by region, depth or language (blocks 08–10 also have Czech tellings), and games. Every block has a ✍️ button that opens the **author studio**. A draft shared with the group code `polirural` appears in the consortium gallery (`/#trida-polirural`), where it collects comments and votes. Editors then **lift** the best drafts into the served book, and the author's name travels with the block.

**For AI agents**, the same content is exposed without a browser:

- **static pages under `/kb/`**, readable without JavaScript;
- **`/llms.txt`**, a short map of the book for language models;
- the **raw markdown** of every block, frontmatter included;
- **open data files** (pilots, practices);
- an **MCP endpoint at `/api/mcp`**, with tools to search the book and list pilots and practices.

MCP is the protocol JackDaw and Raven already use (D4.2), and JackDaw lets users connect their own MCP servers at runtime. In principle, the book can therefore become one more tool in a JackDaw conversation.

## Why bother

The pilots' policy brief (D7.3, Annex 10) calls the human effort of curating local data the **"hidden infrastructure"** of rural AI: months of manual work that is rarely funded. A book makes that work visible, attributed and reusable. D4.2 sets the matching goal for the platform: resources usable by API workflows and AI agents, not only by graphical interfaces.

Success means a reader, human or machine, can move from a practical question to a place-specific explanation, inspect the evidence, suggest a correction, and [reuse the pattern](#c/replicate-local-kb) in another region.

**Sources:**
- D4.2 PoliRuralPlus platform Beta, §2.4 Data Spaces, EOSC and external infrastructure alignment; §4.4 Provenance, data quality and reuse conditions; §3.3.1 Extensibility via Model Context Protocol — [PDF](https://www.poliruralplus.eu/docs/421/D4.2_PoliRuralPlus_Platform_Beta.pdf)
- D7.3 Policy Brief No. 1, Annex 10: Execution Challenges Observed Across Pilots — [PDF](https://www.poliruralplus.eu/docs/349/D7.3_Policy_Brief_No.1.pdf)
- This book's machine brief and map: `/AGENTS.md`, `/llms.txt`