Local Markdown workspace · Claude Code + Codex

The workspace your AI reads before it answers.

Phronesis brings forward the decisions, research, and context that matter before your AI answers. You approve what becomes current, and it captures what changed after the work—all as plain Markdown on your own disk.

See what it writes

Illustrative example
Source note

“Three customers said the address form stopped them from checking out.”

Approved by you
Current decision Approved
Decision

Fix checkout first.

Because

People cannot complete a purchase today.

Returns

When you plan next week’s work.

domains/business/compiled/decisions/2026-07-15-fix-checkout-first.md

Source caught Human gate Used now
Claude Code
Shipped. Works from the same local workspace.
Codex
Shipped. Open the same domain workspace in the CLI or Codex app.
Cursor · Claude Cowork · what ships next
Portable by design: plain files, no runtime required to read them.

Your prior thinking, when you need it.

You bring the moment. It brings the rest. A small, sharp center that perceives, captures, recalls — and carries the judgment to bring the right prior forward. Not a feature list; a presence that has been paying attention.

You make a hard call on Tuesday. By Thursday you would have forgotten the reasoning. The workspace caught it. When the next call rhymes with the last, your thinking is already there.

You do not have to remember to ask. The system has been listening.

Your data. Your codex. Your moat.

The workspace is a directory of structured Markdown files on your own disk. Export it whenever you like. Run any model against it. Move to a different surface tomorrow and the codex/ you have built is still yours.

Phronesis does not own your work — it is the means by which you own it. The architecture is engineered against lock-in, not toward it: export cleanly, swap models, run your skills outside the runtime that made them. Open the same files in Obsidian or VS Code with zero Phronesis dependencies.

The substrate beneath whatever surface you use.

Not another chat window. Phronesis is the compounding workspace underneath the AI tools you already reach for — three layers, two gates, where nothing graduates silently and nothing canonicalizes without your review.

raw/
Append-only source material. Captures and ingests, untrusted by default. Where everything begins.
compiled/
Curated current state, per domain. What is true now. The compile gate sits before this layer.
codex/
Installation-shared, durable principles. What survives across domains, and travels with you.

Bring your own surface. The substrate is what keeps, what curates, and what comes with you when the next model arrives.

The workspace your agent will need.

Today's agents act from thin context. They make plausible-looking calls anyone with real context would catch. The same workspace that catches your judgment is the workspace an agent can act from when you delegate.

You build it once, for yourself. When the agent era lands at your desk, the question stops being whether the model is smart enough. It becomes whether the agent has anything worth working from. The substrate is the answer.

What Phronesis will not do.

  • We won't lock you in.

    Your workspace is Markdown on your own disk. Export it whenever you like, and run any model you choose against it.

  • We won't be a closed-loop vault.

    What you build, you take with you. Phronesis is the means by which you own your work, never its destination.

  • We won't dress rule-following as judgment.

    Frameworks inform the call. They are not the call. Phronesis is built for the situation no rule fully captures — the part that needs the trained eye.

The judgment is yours. We make it compound.

The moat is yours, not ours. Whatever frontier model arrives next is just better fuel for the workspace you have already built.