@indexyz/pi-provider-sub2api@0.3.1

Patch Changes

• Raise the native Codex compaction timeout from three to five minutes. Sessions holding more than roughly 200k tokens regularly need longer than the old limit for a remote compaction round trip, and the abort threw away the native checkpoint and forced the textual fallback.

@indexyz/pi-autocompact@0.2.0

Minor Changes

• Save a model's compaction threshold with /autocompact 300k --save so later sessions reuse it. The value is stored in user settings under piAutocompact.modelLimits. Without --save, the threshold stays session-only.

@indexyz/dsh-autocompact@0.2.1

Patch Changes

• Fix activation on dsh 0.2.1, where the preset registry no longer exports serviceForAgent. Preset compaction backends are resolved with agentPresets.serviceFor, which 0.2.0-rc.2 and 0.2.1-rc.2 both provide.

@indexyz/pi-provider-sub2api@0.3.0

Minor Changes

• Add Sub2API Grok entries with api: "xai": Grok chat discovery with media filtering, xAI catalog metadata and reasoning-effort bounds, encrypted reasoning replay, local function tools kept local by opting out of the Free-account mixed-tool cache route, and Pi textual compaction. Configuration errors for invalid base URLs and failing !command tokens now name the provider and field and no longer echo the configured URL. Breaking: the extension now requires Pi 1.4.0 or newer.

@indexyz/pi-provider-sub2api@0.2.0

Minor Changes

• Fast, Ultrafast, and Daybreak are now per provider and model: requestModes and modelRequestModes in sub2api.json set saved defaults, toggles change only the selected model in memory, and the new /sub2api-save (or /sub2api-save provider) writes the current modes to the configuration. Breaking: toggles no longer apply to every provider and model, and turning a tier off now sends service_tier: "default" instead of leaving the request tier untouched. Native compaction without a configured tier follows the conversation model's resolved tier.

Patch Changes

• Fix native Codex compaction loading in bundled Pi installs.

@indexyz/dsh-prompt-suggestions@0.1.0

Minor Changes

• First release: show a dim next-message suggestion in the empty dsh composer after a successful live reply, accept it with Tab or Right Arrow without auto-submitting, dismiss it with Escape, and persist a General settings switch. Generation uses the conversation's current model with low reasoning, at most eight recent visible exchanges, a 30-second deadline, no tools, no session writes, and no goal inheritance.

@indexyz/dsh-custom-provider-reasoning-efforts@0.2.0

Minor Changes

• Declare vision modalities for hand-written provider routes, and discover a bare
  gateway inventory's model metadata (context window, output cap, display name)
  without overwriting authored values.

@indexyz/dsh-autocompact@0.2.0

Minor Changes

• Add a deployment-wide default threshold: set it in Settings → Autocompact and every model without its own threshold compacts early at that count, while per-model entries and /autocompact still take precedence. Leaving it empty keeps those models on the backend's own compaction.

@indexyz/dsh-thread-summary@0.1.1

Patch Changes

• Release the complete session-title plugin with configurable summary models, native DSH settings controls, bounded generation, deterministic fallback and manual-rename protection. Verify the installed package and browser module for the GitHub Actions release.

@indexyz/dsh-custom-provider-reasoning-efforts@0.1.1

Patch Changes

• Read a Codex-shaped relay's reasoning levels from its /backend-api/codex/models manifest when its /v1/models inventory publishes none.