CtrlK
BlogDocsLog inGet started
Tessl Logo

jbaruch/koog

Koog 1.0 idioms, gotchas, and scaffolding skills for Kotlin agents on the JVM

87

1.85x
Quality

88%

Does it follow best practices?

Impact

87%

1.85x

Average score across 45 eval scenarios

SecuritybySnyk

Advisory

Suggest reviewing before use

Overview
Quality
Evals
Security
Files

task.mdevals/scenario-1/

Ground an Agent in a Documentation Corpus

Problem/Feature Description

A developer maintains an internal documentation site (~2000 pages of markdown). They want a Koog 1.0 agent that answers questions about those docs by retrieving relevant pages before responding. They've already loaded the markdown into a List<Document> where each Document has an id, text, and path.

The agent uses OpenAI. Retrieval should happen when the LLM decides it needs context — they specifically don't want every input augmented with retrieval results (some queries are conversational and don't need docs).

Output Specification

Walk through the wiring — embedder, vector store, retrieval surface, agent integration. Produce the relevant code as a single Kotlin file, labeled.

evals

scenario-1

criteria.json

task.md

README.md

tile.json