Context, memory and models

Headroom shortens material before it reaches a model. Start with a fictional long log so you can check whether a known error survives.

Selecting skill text based on repeated task results.

Task templates for different models and tasks.

Shorter answers and context handling.

Settings for Claude Code from publication.

Compare DeepSeek Harness and Codex on one practice task. Harness runs a separate agent environment with its own model and access settings.

Voice Input in Mac Applications.

Local request gateway to selected providers.

Routing requests through a single local address.

Memory between work sessions in supported environments.

Apply for a single measurement of the local model. Check hardware support, disk, memory, and speed.

Request classifier. Measure selection error and uncertainty class, then delay.

A Markdown notes editor. Links and search help organise files; creating a folder does not automatically enable AI or sync.