dsh-context-budget

Hand-declared third-party gateway routes rarely match the built-in pi-ai catalog, so the context window silently falls back to the 262144 default; meanwhile the compaction backend only accepts a ratio, leaving no direct way to express a compact-at-600k trigger. dsh-context-budget consolidates these scattered settings into a single field group on the Settings → Models provider card: context window, compaction threshold, retain budget, and thinking effort. Fill them once and the values hot-apply in about a second with no second restart.

Core capabilities: the plugin converts your absolute threshold into the ratio the backend expects and writes it to cordis.patch.yml; it places a thinking-effort picker at the bottom-right of the composer, auto-matching vendor profiles for xAI, OpenAI, Anthropic, DeepSeek, Google, and others; every write is validated for correctness, preserves all unrelated lines in the file, and rolls back immediately on failure with a backup left behind.

Ideal for developers running pi-ai routes through self-hosted or third-party gateways who need precise control over context capacity and compaction triggers, and who want to manually set reasoning-effort levels for models from different vendors.