Sage2 by Levanto Labs features an open-source research harness and decision model designed to interpret application screenshots and execute keyboard and mouse controls.
What the product does and how it is positioned
Sage2 by Levanto Labs is built around Sage, a decision model that provides frontier-model judgment in milliseconds accompanied by calibrated confidence scores. The accompanying open-source harness, SageCraft, demonstrates how decision models can evaluate graphical environments and issue control inputs based on visual and log data.
The system operates entirely through visual and standard input channels, utilizing screenshots and combat logs rather than memory injection or modifications. Designed as a research demonstration under the Apache 2.0 license, it includes tools for replaying recorded session events locally without requiring live application connections.
Source-supported ways to use the product
The official demonstration presents a use case where the harness guides a character through gameplay actions, enemy selection, and healing based on visual screenshots.
Developers and researchers can load recorded session events into a local SQLite database to review execution summaries without needing an active API key or game client.
The documented workflow, where available
The harness captures an image of the live application window, including target close-ups and player frames.
The system structures the visual data into a prompt question for the decision model.
The Sage model evaluates the visual data and selects a corresponding action from available options.
The harness validates the chosen action and translates it into standard keyboard or mouse presses.
SageCraft functions as an open-source harness operating under the Apache 2.0 license. Rather than interacting with internal application memory or bypassing security protections, it relies strictly on external observation through application screenshots and ordinary input simulation.
The integration demonstrates how decision models handle conditional branching using calibrated confidence scores, returning null responses when certainty falls below operational thresholds to allow graceful fallback handling.
Checks to run with your own material and workflow
What was checked and when
Answers based on the source-checked product record
SageCraft is an open-source harness under the Apache 2.0 license that connects the Sage decision model to interactive application windows using screenshots and ordinary input methods.
No, it does not read or modify game memory, inject code, or circumvent anti-cheat measures, operating solely through visual screenshots, combat logs, and standard input.
The model supports five answer shapes including yes/no, choice, scale, tags, and sort, delivered from a single API alongside a 0 to 1 confidence score.
Yes, recorded campaign events can be loaded into a local SQLite database to print a summary without requiring an active game client or API key.
No, SageCraft is not affiliated with, endorsed, or sponsored by Blizzard Entertainment, and automated play may violate specific terms of use.
