What the AI OS is
What the AI OS is: the four components of the operating environment, the invariants that bind them, and what the system is not.
The AI OS is the personal operating environment we use to work with AI agents. Its name is Ordova. Four components, each with a single load-bearing job, intended to be portable across any agent and any machine.
The four components
- Charter: hand-authored prose distilling cross-cutting principles, role postures, and discipline. The agent-agnostic, durable layer. It lives as markdown anyone could read and doesn't depend on a specific tool.
- Executive: the executable layer that puts the prose into practice on a machine. Hooks, skills, agents, and settings; Claude Code-specific today, designed to swap.
- Archive: an Obsidian knowledge vault holding long-running knowledge: programme docs, daily notes, handoffs, captured ideas, references. The persistent state the agent reads from and writes to.
- MCP server: the API surface that lets agents query the archive programmatically (search, read, append). The interface the executive depends on but the charter does not.
The charter is plain markdown and could be read by anyone. The archive is private working state, read by us and by the agent when invited via the MCP server. The executive only runs where Claude Code does. The MCP server runs wherever we point it.
What makes this an "OS"
The components share a small set of cross-cutting invariants: the rules that hold across every component, irrespective of which one is changing.
- Central source of truth, derived state. Each fact lives in exactly one place; every other surface that asserts it is a derived view.
- Synthesis is a derived view. Rollups, indexes, and summaries are never authoritative. Change the canonical, regenerate the rollup.
- Portable, not provider-locked. The charter travels with us to any agent; the executive is the only agent-specific layer.
- Discipline encoded twice: once as prose, once as enforcement. The charter names the rule; executive hooks or audits catch when it's violated.
The combination of these invariants, across charter, executive, archive, and MCP, is what makes the system an OS rather than just a directory of files.
What it is not
- Not a product. No-one buys the AI OS. It is professional capital, kept private by deliberate decision.
- Not a framework anyone else uses. The patterns here may generalise; the specifics (our role, our projects, our machines) do not.
- Not Claude Code. Claude Code is the current executive host. The AI OS would survive a swap to another agent runtime; only the executive layer would need replacing.
- Not the archive alone. Many systems are "an Obsidian vault and some scripts". The charter and executive layering is what distinguishes this from that.
Where to read next
- Why it exists: the problem this system solves
- Design philosophy: the central-source-of-truth principle as it applies across the four components
- Operating philosophy: the solo-studio postures the executive encodes
- Methodology: how docs, decks, and derived views are produced and kept in sync