Trust
Craft Phronesis is a folder of Markdown files on your disk. The disk is the source of truth at every tier, and this page is the exact accounting of what we host, what we can see, and what touches a model.
Your local folder is the source of truth.
Your installation is a directory of Markdown files on your machine: your domain workspaces and your shared codex. No database, no proprietary block format, no upload step. The folder is the source of truth.
This holds at every tier. On Free, nothing is hosted. You run the command-line tool, your own AI surface, and your own connector bridges, and the whole installation sits on your disk. On Pro, the hosted pieces are conveniences over that same folder: an indexer so you never configure search, click-to-connect connectors, cloud sync across your own devices, hosted hooks, and a browser onboarding wizard. The cloud copy is a replica. The local folder stays the source of truth. Turn Pro off and nothing essential is lost.
No telemetry in V1.
Nothing, by default and by design. Phronesis ships no telemetry in V1: no analytics, no usage pings, no opt-in metrics. The event log that records what happened inside your workspace is not emitted as analytics or usage telemetry. Connectors pull content in and never re-emit it to any third party. The installation is a sink, not a hub.
On Pro, cloud sync replicates your files across your own devices, and the local folder stays the source of truth.
The model is your choice.
Phronesis does not run a model of its own. It reads and writes your local files and hands the reasoning to the AI provider you choose. On Free, that is your own API key or your own AI subscription: your content goes to the provider you already use, under your own account, and Phronesis is not in the path. On Pro, you can keep bringing your own key, or use an optional bundled AI option where Phronesis fronts the provider relationship for you. The model is a single config value, and you can change it at any time.
Ownership you can check.
Ownership you cannot check is a slogan. Three tests make ours checkable, and V1 cannot ship with any of them failing.
Export: the command that packages your installation must produce a folder that opens cleanly in Obsidian and VS Code with zero Phronesis dependencies present, and keeps working with the vendor gone.
Model swap: changing a single config value to a different provider must preserve the provider boundary the workspace runs through, so your work never depends on a single AI vendor.
Skill portability: a Phronesis skill must run under raw Claude Code, Cursor, or Codex with no Phronesis runtime present.
If any of the three fails, the release does not ship. The ownership is tested in the same place the rest of the product is tested, or it is nothing.
Employer content stays behind the gates.
If you are a PM, the content you most want Phronesis to hold is often your employer's: meeting transcripts, stakeholder threads, calls made in rooms you were in. Here is the honest picture.
Everything not authored by you is treated as untrusted, including meetings you attended. Your presence in the room does not sanitize what other people said in it. Untrusted content lands in your workspace's raw folder and never becomes canonical state on its own. Turning it into compiled state always passes through a review step you approve. Nothing you bring in is re-emitted to a third party.
What Phronesis cannot do is know your employer's policy on where that content is allowed to live. Because your workspace is a local folder and your model provider is your own choice, you keep the controls that answer that question. Check your own agreements before you point a connector at work content. This is a description of how the architecture behaves, not legal advice.
The short version: the disk is yours, the model is your choice, and the ownership is tested, not asserted.