Changelog

1.0.5

Require explicit confirmation before token-consuming provider tests. Track API versus manual-test origin in usage history and CSV, with an additive migration for older records.
Mark estimated totals consistently and explain idle traffic, metadata requests and quota reservations in German and English.

Make confirmed provider deletion available on Overview as well as Providers; remove deleted profile references and empty routing rules.
Integrate the API-key visibility control as an eye icon inside the input, with state-aware German/English accessible labels and tooltips. Preserve automatic secret-reference visibility.

Replace the browser-native model datalist with an explicit searchable dropdown attached to the same model input. Opening the dropdown shows all loaded models without deleting the saved model name; supports mouse and keyboard selection.

Fix automatic provider-model discovery being blocked by failed or slow credential prefill and SDK model catalog requests. Keep saved credentials server-resolved and prevent late SDK suggestions from replacing live results.
Remove the forum-post draft from the repository.

Prefill the original API credential entry on profile editing through a dedicated ingress-only endpoint; recommend secret references and offer a literal-key reveal control.
Add read-only OpenRouter/DeepSeek account metadata, documented plan hints and optional model-specific quota examples.
Document concrete alias/tool/catch-all rule examples in German and English.
Switch app versioning to MAJOR.MINOR.PATCH and mark the app stable.
Show providers in numbered vertical lists with drag-and-drop and accessible move buttons, saving order immediately.
Keep !secret references visible while typing their prefix; restore password masking on any mismatch.
Automatically fetch provider models on editor opening and completed credential changes; show suggestions directly in LLM model with SDK/manual fallback, removing the separate load button and picker.
Add navigation, provider and action icons and color the footer heart red.
Expand German and English READMEs to cover installation, all provider templates, configuration, quotas, routing, Assist/MCP and operational limits.
Link the updated developer credit to the project repository.
Add a More Home Assistant Apps button beside Support my work ☕ in the ingress footer.

Pause and resume provider profiles directly from overview and provider cards, preserving credentials, priority, rules and usage. Paused profiles stay paused across restarts until resumed; already running requests finish normally.
Add a German/English setup path for the HACS MCP Assist integration: HACS repository link, integration form, correctly normalized OpenAI-compatible server URL, router credential and model/timeout copy fields.
Document MCP Assist installation, entity permissions, token-saving settings and troubleshooting without editing HA files or sending credentials through links.
Load model lists directly from supported providers using draft API keys or saved secret references. Add a provider-result picker, bounded pagination and clearly separate SDK suggestions; never follow redirects with credentials.
Add the requested developer credit and PayPal support link to the ingress footer.

First experimental Assist AI Router app, powered by LiteLLM 1.106.4.
Ingress UI in German and English, 21 provider templates, editable models and ordered profiles.
Priority, quota, round-robin and local-first policies; custom alias/tool rules.
Shared-account quotas, atomic reservations, persistent cooldowns and per-attempt token usage.
Supervisor option synchronization with preserved !secret references and conflict detection.
Guided native Home Assistant LiteLLM/Assist and optional MCP setup.
Separate authenticated chat API, buffered SSE and opt-in stable-release checks.