I’m the developer of Read19, an independent AI Go review app. I recently added a workflow for OGS players and would love feedback from people who review their games here.
With Read19, you can:
paste a finished OGS game link, or enter a public OGS username and choose from recent finished games;
run KataGo analysis in the cloud — no engine, weights, or local GPU setup required;
read win rate and score lead together, jump to turning points, and explore candidate moves and playable variations;
use the same review workflow on the web, Windows, Android, and macOS.
Read19 only reads public OGS game history and never asks for your OGS password. For a private or hidden game, you can export the SGF from OGS and import that file instead.
Try the OGS import and review workflow:
For pricing transparency: Read19 is currently in free beta with no card required, and registered accounts currently receive 100,000 KataGo analysis units each month. Paid subscription and one-time analysis-unit packs are also available for higher usage.
Read19 is an independent third-party app, not an official OGS or KataGo service.
I’d especially appreciate feedback on:
whether importing by OGS username or game link feels smooth;
whether the turning-point workflow helps you understand a game;
which OGS-specific review features you would like next.
Thanks — I’ll be following this thread and responding to feedback.
How can i input my ogs username and have it analyse my game history? I can’t seem to find that option >___>
edit: ah nvm, i manged to find that eventually
edit2: i feel like its giving me too much info tbh.. I don’t disargee with it, afterall we could have played much better on the game i tested, but i also feel like these opening moves / joseki options were not all that important for the final reasult:
I would claim that we just played a joseki, that was sub-optimal considering the overall board position, but it was probably not the most important thing to learn from this game
Thanks — this is very useful feedback, and I agree with the distinction: a move can be objectively suboptimal without being one of
the most important lessons from the game.
I found the review shown in your screenshot (OGS game 88770631) and have updated Read19 based on your feedback:
OGS imports now remember which color belongs to the searched player, including Rengo games.
“Key moments” now defaults to showing only your own moves, with an option to view both players.
Problems are sorted by severity first, followed by score loss and win-rate loss.
The initial view highlights up to three distinct learning moments. When several issues occur within four moves of each other,
only the most important one is highlighted, while the complete list remains available.
I also rescanned the existing OGS reviews, so your saved game is now correctly identified as you playing White.
The update is live now. Please refresh Read19 and reopen the game — I’d be very interested to know whether the review now feels
more focused and useful.
Thanks again for taking the time to explain the problem so clearly!
Some it suggests the same move that was played in the game
The matching position search doesnt seem to work, or it uses some very limited pool of games. It finds just 3 games where black opens at Q16, and no games where white plays D17 for the move 2.
What does “Instinct 38.8%” mean?
These sentences are weird: “Compare ownership after the game move and the recommendation to see where this move mattered.” “Reopen an interrupted review to fill in missing positions automatically.”
Thanks — all of these were good catches. I have shipped another update based on them.
OGS reviews now use the imported account player’s perspective throughout the board, candidates, variations, and graph, so your White game is no longer shown from Black’s perspective.
The graph no longer prints loss numbers over the line. Hovering or dragging now gives a fixed readout, and marked mistakes have a tooltip with the move, win rate, score, severity, and loss.
If the played move is also KataGo’s #1 recommendation, it is now always 0.0 points / 0.0% loss and Stable. The bad label came from comparing two separate root searches; the comparison now uses the candidate from the same pre-move search. Previously saved reports with this inconsistency are repaired when loaded.
I removed the Instinct / model-evidence box and the initial-preference values from the UI. That value was KataGo’s policy prior: it was not an earlier WSS update, and it was not Human SL. For a position analysis, WSS streams newer tree-search results until the requested visit limit is reached. A whole-game review finishes the positions, then automatically runs a deeper second pass on selected key positions. The UI now focuses on the latest tree-search result; Human SL remains a separate optional prediction.
You were right about matching positions: it currently searches only games analyzed and synced with Read19, including rotations/reflections, not the full OGS archive. The UI now says this explicitly, so three Q16 games is a coverage limitation rather than a claim that only three exist on OGS. Full-archive matching is not implemented yet.
I also rewrote the ownership and interrupted-review explanations in plainer language.
These changes are live now. Thank you again for the detailed screenshots — they made the problems much easier to identify.
I checked out analysis for one of my recent games. It was perfectly fine.
I guess what I’m mostly wondering at this point: in the last month I’ve tried maybe a dozen new services that all offer to use cloud-hosted Katago to analyze my games with a clearly LLM-generated UI that is mostly just naively rendering Katago output in a similar way as every other tool. It’s not clear to me why I’d choose this one over any of the others.
Honestly, I agree with the underlying criticism. The web UI is only a convenience layer, not the reason I expect people to choose
this service. I’d recommend using the WSS API with the Go software you already like.
What I’m trying to offer is inexpensive, broadly accessible KataGo compute—not another opinionated game-review UI. I think KataGo
analysis should be cheap enough to be available everywhere.
