SPECIFICATION

MECHANICS

1

Memory Index receives objective and referred model API, opening browser instance.

2

Memory Index processes live content for model legibility, supplementing prompt with previous similar actions and multi-step supervision.

3

Model and Memory Index enter into iterative loop to achieve objective.

COMPOSABILITY

High Dimensional Research products can be split apart and composed to best suit your use case, if you prefer not to use a consolidated endpoint.

Memory Index can be connected independently of its browser instance, enabling user browser navigation.

Our passthrough browser instance can be driven by your model without Memory Index for configuring specific use cases without wrangling headless libraries.

Models connected to Memory Index can include specific providers or your own model endpoint.

IMPROVEMENTS

Across sessions Memory Index gradually forms "muscle memory" around the passthrough model, informing page navigation for specific objectives.

Collective sessions and user-defined teacher forcing for specific tasks continually improve the success rate for autonomous tasks.

We employ well-defined schematics for assessing task performance, grouping by type via WebArenas.

Read our blog post on WebArenas and assessing autonomous agent performance on the web

STATUS