The convergence story finally has dollar signs attached.

What's new

• Token usage tracking. Every LLM batch now records `prompt_tokens` + `completion_tokens` from the response. Surfaced in Stats:
  • Tokens (today / week / all time)
  • Daily cost sparkline (last 7 days, in cents)
• Cost estimation. Set your model's per-1M token prices in Settings → LLM → Pricing and Stats shows estimated USD cost per day / week / total.
• Auto-fill from model. Click the link next to the Pricing fields and the extension prefills public list prices for known models (DeepSeek, GPT-4o family, Claude Haiku/Sonnet/Opus). You can override anything — these are ballpark, not gospel.
• Pricing is optional. Skip it and Stats still shows token counts, just no $.

Why

"Convergence" is the central pitch but it was abstract — "LLM calls drop over time." Now you see it as $0.001 today trending toward zero, with a sparkline that flattens. That's the part that actually matters.

Example (DeepSeek-Chat at $0.27 / $1.10 per 1M)

A typical batch (~3K in, ~500 out) costs roughly $0.0014 — about a tenth of a cent. As your library grows and the local matcher catches more, even that fades to nothing.

Install

Download `xspamcast-0.3.0-chrome.zip` below and reload the extension. Existing Library / Whitelist / Settings preserved.