Ontos is a local-first AI context management system. It provides a repo-native knowledge graph built in markdown and YAML, designed for inspectable, repeatable workflows across AI tools.

The Problem

Most AI context solutions treat knowledge as something to be retrieved — embedded, indexed, and searched through opaque vector databases. This makes AI workflows fragile, non-reproducible, and impossible to inspect. You can’t see what the model “knows,” and you can’t verify its reasoning chain.

The Approach

Ontos takes the opposite approach: context should be readable, not retrievable. Every document lives in your repo as plain markdown with YAML frontmatter. Dependencies are explicit. The knowledge graph is a set of files you can open, read, and version-control.

How It Works

  • Documents are markdown files with structured frontmatter (type, status, dependencies)
  • Context maps are auto-generated tiered indices for AI orientation
  • Session logs capture what was done, what was decided, and why
  • Validation enforces dependency depth limits and structural integrity

The CLI (pip install ontos) provides commands for map generation, health checks, document queries, and session logging.