Markdown notes
Plain markdown files you wrote. The easiest place to edit the graph, and the system never touches them. On your machine, in your folders, yours all the way down.
Legal
What's stored where, what's regenerable, and what never leaves you. Your knowledge graph is yours all the way down.
From the longest-lasting to the easiest to rebuild. The higher up, the more it's yours; the lower down, the more it's a regenerable cache. Concepts and relationships found in your markdown become the graph, and everything below it is regenerable.
Plain markdown files you wrote. The easiest place to edit the graph, and the system never touches them. On your machine, in your folders, yours all the way down.
Portable, standard-format knowledge graph files that sit next to your notes. Built from the concepts and relationships found in your markdown. You can diff changes at a glance, and they're rebuilt from your writing anytime.
Topic reports built by automatically clustering related subjects. Plain text that both humans and AI can read. Regenerable too.
An index that learns the terms you use in your notes and is optimized for fast search. A cache that makes search fast. All rebuilt from your notes.
The most volatile layer, there to keep search and connections running fast. Delete it and it's restored from your knowledge graph files and notes.
Four principles for how we handle your data.
The knowledge graph, search index, knowledge graph store, and topic maps are all built from your markdown. Delete them and they grow back from your writing. The graph never holds your words hostage.
Reading the concepts and relationships out of your notes, and turning a query into a vector, are AI calls. That text travels through our metering proxy — which relays it without storing the content, keeping only usage metadata — and on to the model provider that runs the AI. Ranking, the knowledge graph, and keyword search run on your machine, and your notes, the graph, and the search index never leave it. Web access is off by default.
We don't train AI on your data.
Knowledge graph files are a portable, standard format, and markdown opens in any editor. Nothing locks you in. Delete Consilience and your notes and graph stay right where they are.
Consilience is local-first. Your knowledge lives across several layers, but the order is clear. Concepts and relationships found in your markdown become the knowledge graph, and everything below it is a regenerable cache. In order of how long it lasts:
Everything below markdown notes is regenerable. The knowledge graph, topic maps, search index, and knowledge graph store are all built from your writing. Delete them and they're rebuilt from your notes. The graph never holds your words hostage. Delete Consilience and your markdown stays right where it is.
Your notes, the knowledge graph, and the search index live on your machine and stay there. Two things do leave: the note text we read concepts and relationships out of, and the text we turn into vectors. Both travel through our metering proxy — it relays them without storing the content, and keeps only usage metadata like the model and token counts — and on to the model provider that runs the AI. Ranking, graph traversal, and keyword search are local. Web access is off by default, and untrusted web content is handled separately.
We don't train AI on your data. We never intercept, retain, or use your notes for training.
Nothing locks you in. Knowledge graph files are a portable, standard format, and markdown opens in any editor. You can take your notes and knowledge graph and leave anytime. That's the local-first promise.
Local-first, no training. Build the graph with markdown, and the agent reasons on top of it.