> ## Documentation Index
> Fetch the complete documentation index at: https://docs.leeroo.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Overview

> How domain knowledge is acquired, stored, and used

## Overview

The Knowledge System provides domain-specific guidance during problem-solving. It helps the agent recommend proven approaches and avoid common pitfalls by learning from repositories, research, and past experiments.

## Key Concepts

### Knowledge Graph (KG)

The KG stores domain knowledge in a structured format:

* **Vector Storage**: Weaviate for semantic embeddings and similarity search
* **Graph Storage**: Neo4j for relationships and graph traversal
* **Content**: Wiki pages with overviews, detailed content, and connections

### Learning Pipeline

A two-stage process for acquiring knowledge from sources:

1. **Stage 1 (Ingestors)**: Extract structured wiki pages from sources
2. **Stage 2 (Merger)**: Hierarchical sub-graph-aware merge into the existing KG

### Knowledge Search

Hybrid retrieval that combines multiple approaches:

* **Semantic search**: Vector similarity in Weaviate
* **Graph traversal**: Connected pages from Neo4j
* **LLM reranking**: Relevance scoring for better results

## Using Knowledge in Kapso

### Option 1: Pre-indexed KG

```python theme={null}
from kapso.kapso import Kapso

# One-time: Index wiki pages
kapso = Kapso()
kapso.index_kg(
    wiki_dir="data/wikis_llm_finetuning",
    save_to="data/indexes/llm_finetuning.index",
)

# Every time: Load existing index
kapso = Kapso(kg_index="data/indexes/llm_finetuning.index")
solution = kapso.evolve(goal="Fine-tune LLaMA with QLoRA")
```

### Option 2: Learn from Sources

```python theme={null}
from kapso.kapso import Kapso, Source

kapso = Kapso()

# Learn from a repository
kapso.learn(
    Source.Repo("https://github.com/huggingface/transformers"),
    wiki_dir="data/wikis",
)

# Learn from web research
findings = kapso.research("QLoRA best practices", mode="idea")
kapso.learn(*findings.ideas, wiki_dir="data/wikis")
```

### Option 3: Research as Context

```python theme={null}
from kapso.kapso import Kapso

kapso = Kapso()

# Research without persisting to KG
findings = kapso.research(
    "unsloth FastLanguageModel example",
    mode="implementation",
    depth="deep",
)

# Use research as context
solution = kapso.evolve(
    goal="Fine-tune with Unsloth + LoRA",
    context=[findings.to_string()],
)
```

## Search Backends

| Backend             | Data Format        | Storage          | Use Case                       |
| ------------------- | ------------------ | ---------------- | ------------------------------ |
| `kg_graph_search`   | Wiki pages (.md)   | Weaviate + Neo4j | Semantic search with reranking |
| `kg_llm_navigation` | JSON (nodes/edges) | Neo4j only       | LLM-guided graph navigation    |

## Infrastructure Requirements

Both backends require database infrastructure:

```bash theme={null}
# Start Weaviate and Neo4j
./scripts/start_infra.sh

# Stop infrastructure
./scripts/stop_infra.sh
```

## Next Steps

<CardGroup cols={2}>
  <Card title="Learning Pipeline" icon="graduation-cap" href="/docs/knowledge/learning-pipeline">
    How knowledge is extracted and merged
  </Card>

  <Card title="Search Backends" icon="magnifying-glass" href="/docs/knowledge/search-backends">
    Hybrid search implementation
  </Card>

  <Card title="Configuration" icon="gear" href="/docs/guides/configuration">
    YAML config for knowledge search
  </Card>

  <Card title="Research" icon="globe" href="/docs/research/overview">
    Web research for knowledge acquisition
  </Card>
</CardGroup>