The Kapso knowledge graph: imported domain knowledge

The knowledge graph is Kapso’s imported memory: what other people have written, drawn from repositories and research and searchable during a campaign. It is the counterpart to the lesson bank, which holds what Kapso measured itself, and to Leeroopedia, the hosted knowledge base a campaign reaches with one key.

What is in the knowledge graph?

Wiki pages — overviews, detailed content, and the links between them — held in Weaviate for semantic search and Neo4j for graph traversal. Two subsystems feed and read it:

The learning pipeline turns a source into pages (ingestors), then merges them into the existing graph (a hierarchical, sub-graph-aware merge).
Knowledge search retrieves at campaign time: semantic search, graph traversal, and reranking.

Hosted knowledge: Leeroopedia

The graph is yours to build; Leeroopedia is the hosted layer beside it: a curated knowledge base of ML and AI frameworks that sessions query over MCP with one key in .env and no infrastructure. Ideation asks it how a framework behaves and has its plan reviewed; implementation asks for build plans, documented hyperparameter ranges and failure diagnoses. What each node consulted, and the credits spent, appear in kapso watch --tree and the final summary. See Connect Leeroopedia and MCP gates.

How do I get knowledge into a campaign?

Option 1: Pre-indexed KG

from 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: Ingest sources

from kapso import Kapso, Source

kapso = Kapso()

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

# Ingest web research
findings = kapso.research("QLoRA best practices")
kapso.learn_knowledge(findings, wiki_dir="data/wikis")

Option 3: Research as context

from kapso import Kapso

kapso = Kapso()

# Research without persisting to the knowledge graph
findings = kapso.research(
    "unsloth FastLanguageModel example",
    mode=["idea", "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

What infrastructure does it need?

Both backends require database infrastructure:

# Start Weaviate and Neo4j
./scripts/start_infra.sh

# Stop infrastructure
./scripts/stop_infra.sh

Learning Pipeline

How knowledge is extracted and merged

Search Backends

Hybrid search implementation

Research

Web research for knowledge acquisition

Configuration

YAML config for knowledge search

Related pages: Learning Pipeline · Search Backends · Research · Configuration Kapso is an open-source framework by Leeroo that builds software toward measurable goals through experiment campaigns. Source code: github.com/Leeroo-AI/kapso · Install: pip install leeroo-kapso · Every page as plain text: llms.txt.

Getting started

Evolve system

Knowledge graph

Trajectory learning

Research

Deployment

Benchmarks

The Kapso knowledge graph: imported domain knowledge

What is in the knowledge graph?

Hosted knowledge: Leeroopedia

How do I get knowledge into a campaign?

Option 1: Pre-indexed KG

Option 2: Ingest sources

Option 3: Research as context

Search backends

What infrastructure does it need?

Learning Pipeline

Search Backends

Research

Configuration

​What is in the knowledge graph?

​Hosted knowledge: Leeroopedia

​How do I get knowledge into a campaign?

​Option 1: Pre-indexed KG

​Option 2: Ingest sources

​Option 3: Research as context

​Search backends

​What infrastructure does it need?

​Related

Learning Pipeline

Search Backends

Research

Configuration

What is in the knowledge graph?

Hosted knowledge: Leeroopedia

How do I get knowledge into a campaign?

Option 1: Pre-indexed KG

Option 2: Ingest sources

Option 3: Research as context

Search backends

What infrastructure does it need?

Related