Fathym
Menu

Seeing what your AI knows

Every AI coding tool is a black box. You type something. It writes code. Sometimes it's right. Sometimes it's close but not quite. You have no way to check what context it actually had when it started.

fai --preview opens that box.

One command

fai --preview

This starts fai's MCP server and opens MCP Inspector in your browser. The inspector shows you:

  • Tools - every tool your AI can call (workspace search, session capture, agent coordination, etc.)
  • Resources - FAI-PROJECT.md, FAI-SESSION.md, FAI-PERSONAL.md, petnames, and docs
  • Prompts - the orient, capture-guide, and holonic development prompts that guide your AI's behavior

You can browse entries, call tools yourself, and verify exactly what your AI reads before it writes a single line.

What to look for at Session 1

After your first session, preview will show:

  • A mostly empty FAI-PROJECT.md (fai just started learning)
  • Your first captures from the session hooks
  • All 17 tools available but no crystallized patterns yet

This is your baseline. Everything grows from here.

What to look for at Session 20

After twenty sessions, the same preview shows a different picture:

  • FAI-PROJECT.md is rich with decisions, patterns, and conventions fai extracted from your work
  • FAI-SESSION.md carries forward from where you left off
  • Capture volume shows the depth of context your AI works with

The difference between Session 1 and Session 20 is the difference between a stranger and a colleague who knows your codebase. Same tools. Richer context.

Using preview as a trust-building tool

Preview is useful when:

  • You suspect your AI is missing context (check what it actually has)
  • You switch agents and want to confirm context carried over
  • You want to show a teammate what fai captures about a project
  • You're debugging why the AI made a surprising choice

The inspector is read-only by default. Browse freely. Nothing changes in your vault.

On this page