Kogwistar LLM-Wiki

A continuously-learning knowledge system built on Kogwistar.

Feed it documents. It extracts entities, promotes knowledge, distills wisdom, and projects an interlinked Obsidian vault — automatically, in the background.

raw sources → kg-doc-parser → conversation graph → promote → knowledge graph → Obsidian vault
                                                                               → wisdom engine

What it is

It is not a chatbot, a note app, or a RAG wrapper.

Installation

Option A — Local development (recommended)

bash scripts/bootstrap-dev.sh

The script:

1. Clones kogwistar, kogwistar-obsidian-sink, kg-doc-parser from GitHub if not already present locally
2. Installs all three as editable from the local checkout
3. Installs this package last

After running, the venv always uses local editable sources — re-running is safe (existing checkouts are kept).

Option B — GitHub-only (CI / no local edits needed)

pip install git+https://github.com/humblemat810/kogwistar.git
pip install git+https://github.com/humblemat810/kogwistar-obsidian-sink.git
pip install git+https://github.com/humblemat810/kg-doc-parser.git
pip install -e ".[dev]"

kogwistar and kogwistar-obsidian-sink are not on PyPI — both options install them from source.

Windows: Run the bootstrap from Git Bash or WSL.

Quick demo

# 1. Bootstrap (first time only)
bash scripts/bootstrap-dev.sh

# 2. Ingest a document
python -c "
from kogwistar_llm_wiki.ingest_pipeline import IngestPipeline
p = IngestPipeline(workspace_id='demo')
p.run('my_document.md')
"

# 3. Run the background workers
llm-wiki daemon maintenance --workspace demo
llm-wiki daemon projection  --workspace demo --vault ~/obsidian/wiki

See QUICKSTART.md for the full step-by-step tutorial.

CLI reference

llm-wiki daemon projection  --workspace <id> --vault <path> [--interval <s>]
llm-wiki daemon maintenance --workspace <id>                [--interval <s>]

Full reference: doc/cli_reference.md

Docs

System Documentation

Ecosystem Analysis

Development

pytest tests/unit/          # fast unit tests (in-memory, no services)
pytest -m integration       # Obsidian vault + other on-disk integration checks
pytest -m manual            # opt-in smoke tests requiring local services

After a successful ingest/projection run, the most useful manual checks are:

• maintenance jobs in the durable meta-store should be DONE
• projection jobs in the durable meta-store should be DONE
• the projection manifest row should be ready
• the Obsidian vault should contain the expected .md files