Fathym
Menu
fAIOpenXThe LoopHolonicDocsBlogLog in to OpenX
Docs menu

Your First Session

What Happens When You Run fai

You run:

$ fai

You see (trimmed, with example models):

🧠 Synthesis provider
────────────────────────────────────────
? Select synthesis provider
❯ ollama — qwen2.5-coder:7b (free, local)
  Custom OpenAI-compatible endpoint
  Capture-only (no synthesis)

? Embeddings provider
❯ ollama — nomic-embed-text:latest (768-dim)
  Local ONNX (offline, last resort)
  Custom OpenAI-compatible endpoint

✓ Embeddings configured: ollama — nomic-embed-text:latest

🔧 Your setup
────────────────────────────────────────
  Model:  qwen2.5-coder:7b (local)
  Mode:   watch — fai captures while you code

  Your AI forgets between sessions. fai remembers.
  While you work, fai compounds your context —
  mechanical and AI-driven, growing with you.

That's the only setup. fai asks two questions the first time: which model distills your sessions, and which one indexes them for recall. Every run after that skips them. Change either later with fai synthesize - see Local Models and Tuning.

Leave fai running, and open your AI tool in the same project.

Your First Conversation

In your AI tool (Claude Code, Cursor, ...):

You: I'm building a REST API for a recipe app
AI:  Great. What framework do you prefer?
You: Whatever you think works best
AI:  I'll use FathymAPI - it's lightweight and fast for Deno.
     Starting with the recipe CRUD endpoints.
     [Generates project structure + initial routes]

Notice what happened:

  • You described WHAT you wanted
  • Your AI handled HOW
  • No boilerplate. No scaffolding commands. No decisions you didn't need to make.

Behind the Scenes

While you were building, your workbench was working too:

  • Saved your framework choice (FathymAPI) - won't ask again
  • Noted your style (concise instructions, trust AI's judgement)
  • Recorded the project context (recipe API, Deno runtime)

You didn't tell it to do any of this. It's automatic.

Where fai Makes the Difference

Your first session is roughly the same with any AI tool - you start building.

What changes is every session after the first.

Without fai (Session 2):

You: I'm working on that recipe API we started yesterday
AI:  What recipe API? Can you describe the project?
You: We're using FathymAPI, Deno, and I have 4 endpoints...
AI:  Got it. What patterns are you following?
You: [20 minutes of re-explaining]

With fai (Session 2):

$ fai
? Resume a session
❯ Continue: <your session> · 1 day ago · …
  Start new session
  Manage this session
  Manage Synthesis

Then, in your AI tool:

You: Let's keep going on the recipe API
AI:  Picking up the recipe API - 4 endpoints so far. Add authentication next?

Full context restored. No re-explaining.

That gap widens every session. By Session 50, your AI anticipates your needs. Without fai, you're still re-explaining on Session 50.

See What Your AI Knows

After any session, open a second terminal:

$ fai --preview

This opens MCP Inspector in your browser. You can see every tool your AI can call, every piece of context fai captured, and the full project knowledge fai has built. Browse entries, call tools yourself, and verify exactly what your AI reads before it writes a single line.

First Run vs. Every Run After

First run: fai asks the two setup questions above, then starts your first session automatically. There's no session menu yet, because there's nothing to choose from.

Second run onward: fai opens with a session menu. You can continue a session, start a new one, manage one (rename or finish it), or change your providers with Manage Synthesis. See Sessions.

After Your First Session

Tomorrow, run fai, continue your session, and open your AI tool. No re-explaining. Your AI picks up exactly where you left off. That's the difference.

On this page