Chessy connects an AI agent to a live chess board where a full game can be played from the opening move through checkmate or resignation. The agent submits moves as white on an interactive board, and Chessy's opponent reasons through its position before answering as black, so what comes back is a considered reply rather than a scripted database lookup pulled from a fixed opening book. Through Neotask, an agent can start a game, track the board state move by move, and keep a running log of the position without a person keying anything into a chess app themselves at any point. This works for practicing a specific opening line repeatedly, testing how an agent explains chess reasoning to someone learning the game for the first time, or simply keeping a match going in the background while other unrelated work happens elsewhere in the same session. There is no scorekeeping, coaching commentary, or puzzle library attached to the integration itself, just a real opponent that thinks through the position and responds move by move on the same board the agent is reading from throughout the game.
| Start game | Opens a new match on the interactive board with the agent playing white from move one. |
| Submit move | Sends a move in standard notation and advances the board by one ply. |
| Read board state | Returns the current position in full so the agent can plan its next move accurately. |
| Get AI reply | Triggers Chessy's reasoning pass over the current position and returns its move as black. |
| Resign or end game | Closes out a match cleanly once either side has decided the game is over. |
| Retrieve move history | Returns the full sequence of moves played so far in a given game. |
A user asks their agent to try the same opening five separate times against Chessy and note how black responds differently on each attempt. The agent starts five separate games, plays the identical first four white moves in every one, and records Chessy's replies side by side once each game reaches the same point in the sequence. Because Chessy reasons fresh each time rather than pulling from a fixed book, the five games diverge in interesting ways after move four, giving the user real variation to study instead of a single deterministic line. The agent then summarizes which of the five lines led to sharper middlegame positions for white and which one ran into an early tactical problem worth avoiding. The user ends up with a genuine comparison across five full attempts without having played a single move on the board themselves.
Someone learning chess asks their agent to play a slow game against Chessy while narrating the reasoning behind each move as it happens on the board. The agent submits a move as white, waits for Chessy's response as black, and explains in plain language why that reply made sense given the specific position at that moment in the game. When a move looks unusual to the beginner, the agent walks back through the last two or three moves to show what changed on the board and why Chessy reacted the way it did. Over the course of a full game lasting thirty or forty moves, the beginner watches real, unscripted play unfold with commentary attached directly to each decision. By the end of the game the beginner has a far better sense of how pieces interact than a rulebook alone would have given them.
Chessy's black side reasons through each position in real time before replying, so the moves come from live evaluation of the board rather than a preset sequence or a fixed opening database lookup. That means the same opening can produce a different reply from Chessy on separate games.
The connection is set up for the agent to move as white with Chessy answering as black, matching how the integration is exposed through the interactive board. There isn't a documented way to flip that arrangement within the current integration.
The board state and move history can be read at any point during or after a game, so an agent can log the position after every move and reconstruct the full sequence of play once the game is finished. That makes it straightforward to go back and study a specific stretch of the game later.
An agent can open more than one game and track each board independently, which is useful for comparing lines side by side or running several practice sessions in parallel. Each game keeps its own state, so there's no risk of moves from one match bleeding into another.
The board only advances on a legal move, so an agent working through Neotask should check the current position before submitting a move to avoid having it rejected. Reading the board state first is the simplest way to confirm a move will be accepted.
The integration description centers on live move-by-move reasoning rather than an adjustable difficulty setting, so the experience is built around a genuine reasoning opponent rather than a strength dial that can be turned up or down. The challenge comes from Chessy's actual evaluation of each position rather than a configured skill level.