lorelang

🏛️ Lore

Human-readable ontology format for the AI age.

Define your domain knowledge in prose. Compile to knowledge graphs, AI agent prompts, embedding indexes, or JSON schemas.

Lore is designed for teams that need to encode institutional knowledge — the kind of expertise that lives in people's heads, scattered documents, and tribal knowledge — into a format that both humans and AI can reason over.

Lore is AI-first: unstructured and semi-structured narrative is the primary source of meaning. Structured compilers are best-effort projections of that source.

ontology/
├── lore.yaml              # manifest
├── entities/              # what things exist
├── relationships/         # how they connect
├── rules/                 # what logic governs them
├── taxonomies/            # how they're classified
├── glossary/              # what terms mean
└── views/                 # who sees what

Why Lore?

The problem: AI agents need domain understanding, not just data access. An agent querying a database can retrieve rows, but it can't reason about what those rows mean — how entities relate, what rules apply, or what constitutes a valid chain of inference.

The existing solutions are wrong:

• OWL/RDF: Powerful but unreadable. Nobody outside ontology specialists can author or review them.
• JSON/YAML schemas: Too flat. Can't express inheritance, inference rules, or reasoning paths.
• Markdown docs: Great for humans but no tooling can parse them consistently.

Lore's approach: A markdown-like format with just enough structure that tooling can parse it deterministically, but readable enough that domain experts can author and review it directly. Domain understanding also has to evolve, by humans and AI Agents, Lore comes with opinionated curation workflows for evolution of domain understnading, which can be extended with plugins.

Quick Start

# Install
pip install -e .

# AI-first domain bootstrap
lore setup my-domain --domain "B2B Customer Discovery"

# Validate an ontology
lore validate examples/b2b-saas-gtm/

# Compile to AI agent context
lore compile examples/b2b-saas-gtm/ -t agent

# Compile scoped to a specific team's view
lore compile examples/b2b-saas-gtm/ -t agent --view "Account Executive"

# Compile to Neo4j Cypher
lore compile examples/b2b-saas-gtm/ -t neo4j -o schema.cypher

# Export as JSON
lore compile examples/b2b-saas-gtm/ -t json -o ontology.json

# Generate embedding chunks (JSONL)
lore compile examples/b2b-saas-gtm/ -t embeddings -o chunks.jsonl

# Export interoperable ontology JSON-LD
lore compile examples/b2b-saas-gtm/ -t jsonld -o ontology.jsonld

# Generate Mermaid ER diagram
lore compile examples/b2b-saas-gtm/ -t mermaid -o diagram.mmd

# Generate routing indexes and run quality curation
lore index examples/b2b-saas-gtm/
lore curate examples/b2b-saas-gtm/

# Run self-improving proposal generation from outcomes
lore evolve examples/b2b-saas-gtm/

# Distill a meeting transcript into observations (semi-structured claims)
lore ingest transcript examples/b2b-saas-gtm/ \
  --input ./meeting.txt \
  --about Account

# Distill memory exports into observations (adapters: arscontexta|mem0|graphiti)
lore ingest memory examples/b2b-saas-gtm/ \
  --adapter mem0 \
  --input ./memory.json \
  --about Account

# Review generated proposals (accept/reject)
lore review examples/b2b-saas-gtm/proposals \
  --decision accept \
  --reviewer ontology-curator

# Show statistics
lore stats examples/b2b-saas-gtm/

# Visualize entity graph
lore viz examples/b2b-saas-gtm/

Build Shortcuts

make help
make launch-check
make dist-check

This gives a single command path for test + conformance + example validation + compile matrix + end-to-end smoke checks.

setup also accepts alias forms: /setup, lore:setup, /lore:setup.

File Format

Entity (.lore files in entities/)

---
entity: Account
description: A company that is a customer or prospect.
---

## Attributes

name: string [required, unique]
  | Legal entity name.

health_score: float [0.0 .. 100.0]
  | Composite health metric.
  | @computed: rules/scoring.lore#account-health-score

## Identity

An account is uniquely identified by its primary web domain.

## Lifecycle

Accounts progress through stages: prospect → onboarding →
active → at-risk → churned → former.

Relationships (.lore files in relationships/)

## HAS_SUBSCRIPTION
  from: Account -> to: Subscription
  cardinality: one-to-many
  | An account can have multiple subscriptions.

## Traversal: revenue-by-product
  path: Account -[HAS_SUBSCRIPTION]-> Subscription -[SUBSCRIBES_TO]-> Product
  | Answers: "What products does this account use?"

Rules (.lore files in rules/)

## champion-departure-alert
  applies_to: Account
  severity: critical
  trigger: Champion contact becomes inactive