Skip to main content

Lore

Build and maintain persistent LLM knowledge bases from any content.

Lore solves the "stateless AI" problem -- the context-limit reset where agents lose architectural nuance between sessions. Instead of RAG (vector embeddings, retrieval noise), Lore builds a compiled markdown wiki: an LLM acts as a full-time research librarian, actively organizing and interlinking knowledge.

Quick Start​

npm install -g @telepat/lore
cd your-project
lore init
lore ingest ./docs/
lore compile
lore search "your query"

Continue with:

Learning Paths​

New here​

  1. Overview
  2. Installation
  3. Quickstart

Building and maintaining a wiki​

  1. Ingesting Content
  2. Compiling Your Wiki
  3. Searching and Querying
  4. Troubleshooting

Publishing and sharing​

  1. Exporting
  2. MCP Server
  3. Explain Command

Team operations​

  1. Lore Angela
  2. Best Practices
  3. Linting and Health

Docs Theme Behavior​

The docs site follows your OS/browser color scheme by default (prefers-color-scheme).

What Is New​

  • Compile now uses hash-based incremental behavior to skip unchanged extracted content via manifest.json.
  • Compile now uses a repository lock (.lore/compile.lock) to prevent overlapping runs.
  • Watch mode now auto-compiles raw changes with debounce and queued follow-up pass behavior.
  • Lint now emits line-aware diagnostics in JSON mode alongside orphan/gap/ambiguous summaries.
  • Compile now generates .lore/wiki/concepts.json for canonical concept metadata and aliases.
  • Ingest now auto-normalizes supported chat exports (.json/.jsonl) into transcript markdown.
  • Ingest metadata now includes richer tags (folder-derived and heuristic memory categories).
  • Duplicate ingest short-circuits repeated sources for faster, cleaner pipelines.
  • Index rebuild supports repair mode (lore index --repair) for manifest recovery.
  • Query supports optional typo normalization while preserving technical tokens.
  • MCP toolset now includes ingest/compile write tools, duplicate checks, taxonomy summaries, and lint-focused maintenance tools.

Fast Maintenance Loop​

# 1) Ingest or refresh sources
lore ingest ./docs

# 2) Compile and rebuild index (repair missing manifest entries if needed)
lore compile
lore index --repair

# 3) Validate graph health
lore lint

# 4) Ask questions
lore query "What changed in architecture?"

How It Works​

RAGLore
FormatVector embeddingsStructured Markdown
RetrievalSimilarity searchBacklinks + FTS5/BM25
PersistenceStatelessEvolving wiki + git
MaintenanceManualLLM-driven librarian