AI insights

Phihelm's AI runs before your entry (pre-scan) and afteryour exit (post-trade debrief). It's advisory, never blocking — your trades fire even if AI is slow.

Pre-scan score

Every projected pattern gets one paragraph + a confidence score from −10 to +10. The AI reads:

The pattern itself (type, ratios, ZigZag confirmation).
VIX + breadth + sector rotation context.
Your historical hit-rate on this pattern + symbol combo.
Time of day + day-of-week.

Output looks like: "Pre-scan: TAKE +6 — bullish reversal at VWAP-1σ with breadth flipping bullish (ADD −325 → +120 in 8m). Your FIB_786 longs are 67% over the last 30 days."

Look for

Confidence ≥ +5 with the explanation citing specific indicators (VWAP, breadth flip, money flow). When the AI cites concrete numbers it's reading the data; when it only says "bias is bullish" it's not.

Avoid

Trading against AI confidence ≤ −3. The AI can be wrong, but strongly-negative scores correlate with losses in the backtest. Take the skip; the next setup is always coming.

Post-trade debrief

After every closed trade you get an A–F grade + two paragraphs: what you did right, what you could have done better. The grade weighs entry timing, hold duration, exit decision, and slippage.

A / B — execution matches your edge.
C — the trade worked but the timing was lucky.
D / F — even if you won, you broke a rule.

Tip

Read 50 debriefs in a row and you'll find your own pattern. Most traders bleed on the same 2-3 mistakes; the AI surfaces them in plain English so you can't pretend they aren't there.

Privacy + cost

AI runs on Anthropic Claude by default. Phihelm never sends your broker credentials or unrelated personal data. If you prefer fully local, switch to Ollama in Settings — costs nothing, slightly slower.