What this is
Caissa Research is an AI chess coach that reads your games like a grandmaster — not just an engine that flags mistakes. The product is engine-to-human translation. The game is the vertical; the work is the writing.
How a game becomes a read
What this is not
| Caissa | Not |
|---|---|
| Reads your games like a grandmaster | A puzzle trainer or rating-grinder |
| Engine-to-human translation | An engine UI |
| Coach with memory across games | Per-game tool with no recall |
| Capability-tier pricing | Pay-per-analysis or pay-per-question |
| Multi-source: lichess + chess.com | Locked to one platform |
| Honest empty states until analysis lands | Generated commentary on cold data |
What we mean by "deeper than chess.com"
chess.com Game Review uses Stockfish 16 at roughly depth 20 with one principal variation. We run Stockfish 18 at depth 18 with three principal variations, plus Maia probability modeling on every move, plus deterministic substrate detection across twelve layers. The engine reads further; the substrate explains why.
The premium tier wedge isn't compute volume — it's the writing. A game-review report tells you what was lost. A grandmaster essay tells you what to learn.
Why no compute-based pricing
Most AI products meter you. We don't. The product is a coach with memory; you're buying the relationship, not units of inference. Heavy users aren't the most expensive customers — they're the most engaged. Each tier opens a class of capability, not a higher allowance of the same one.
Status
The roster feed (Magnus + streamers, hourly cron) is live. The substrate layers are shipped. The Kasparov-level prose layer is the active substrate-hardening track. Q&A on a position is the next consumer surface. Editorial-on-demand sits on top of both.
The free tier is real and serving traffic today. Coach and Pro tiers are in active development; the pricing page shows what each unlocks with explicit "soon" badges where capabilities haven't shipped yet.