…plicates

…t type

…codes

Status codes: 正常→active, 限流→limited, 异常→error, 禁用→disabled
Auto-migration in _normalize_account for backward compatibility
All UI text translated: settings, accounts, image pages

… codex chat

…w Vietnamese

…prefetch

…poses

…proxy cards

…se body and headers

…s .text

…llback to Gemini

…tate

… original_token ref

…ffix

…t filters exact match

- openai_default_model config (default: gpt-4o)
- UI input in Settings for custom model name
- Used when chatgpt/auto routes through api.openai.com

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Web session routes through api.openai.com and needs gpt-4o,
gpt-4o-mini, etc. Now always merged into chatgpt provider models.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Added complete guide for new users: HA integration via
Local OpenAI LLM, recommended prompts, model prefixes,
combo fallback, and image generation.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

_ensure_openai_provider() creates the required custom provider
config if missing. No manual setup needed for addon.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

New addon installs now have api.openai.com provider configured
out of the box. No manual setup needed for web session routing.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

When API key is sk-auto-created or sk-placeholder, automatically
uses first active JWT token from account pool. Makes addon work
without manual API key configuration.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Shows token preview, model, message count when OpenAI returns
errors to help diagnose addon 500 issue.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

DATA_DIR auto-uses /data/chatgpt2api when running as HA addon.
Config copied from default on first run. Accounts survive rebuilds.
Debug logs for auto-JWT key injection and error details.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Previously only first JWT token was used, causing "All API keys
rate limited" when that single key hit rate limit. Now all active
JWT tokens are available for rotation.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

gpt-4o returns 500 on free plan JWT tokens via OpenAI API.
gpt-4o-mini is available on all plans.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Remove OpenAI API routing to match original repo behavior.
Fixes JWT token 500 errors on addon (ARM64/IP restricted).

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Previously search was skipped for chatgpt provider. Now
chatgpt/auto uses chatgpt.com backend without browsing,
so search injection is needed for real-time results.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Auto-detect architecture: AMD64 uses OpenAI API (native tools),
ARM64/addon uses chatgpt.com backend (avoids JWT 500 errors).

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

When ARM64 + tools, route via openai_relay_url config.
force_base_url overrides custom provider base URL.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

ARM64 routes: relay if openai_relay_url set, else prompt-based
tool loop. AMD64 unchanged (OpenAI API direct).

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

…64/ARM64 split

- Remove _handle_chatgpt_tool_loop (prompt-based tool calling)
- Remove force_base_url parameter (relay remnant)
- Keep simple: AMD64 → OpenAI API, ARM64 → chatgpt.com backend

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

- Check /data/options.json as definitive HA addon signal
- Works even if /data/chatgpt2api doesn't exist yet on first boot

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

- /config/chatgpt2api survives uninstall, included in HA backups
- Auto-migrate from old /data/chatgpt2api location
- Read-only config map for web UI managed settings

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

- BackendRouter.route(): auto resolution checks model_settings.enabled_models
- openai_oauth: cx/auto fallback chain filtered to enabled models only
- _handle_chatgpt_chat: chatgpt/auto on AMD64 respects enabled filter
- If default model not enabled, falls back to first enabled model

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

…counts)

- _extract_xml_tool_calls_from_text: parse <tool_call> XML blocks from AI text
- _stream_chatgpt_addon: streaming wrapper with tool call extraction
- _chatgpt_addon_completion: non-streaming wrapper with tool call extraction
- Chatgpt.com path in _handle_chatgpt_chat now calls these wrappers
- Shared functions (stream_text_chat_completion, completion_response) unchanged
- No impact on AMD64 OpenAI API path or Anthropic protocol

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Keyword detection for Vietnamese smart home terms injects
MUST call GetLiveContext FIRST hint into system prompt.
Works in both streaming and sync wrappers.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

…I on all archs

Paid accounts (codex, go, plus, business) now route to OpenAI API native
tools regardless of architecture. Free accounts still use chatgpt.com
backend with XML parsing + force hint.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Docker free accounts keep original chatgpt.com backend behavior.
Only addon free accounts get XML parsing + force hint for HA tools.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

enabled_models contains auto entries which are UI placeholders,
not real API models. Filter them out before picking model.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Enabled models have chatgpt/ prefix but default_model is bare name.
Normalize prefix before comparison to correctly detect if default is enabled.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

…tent detected

When needs_search() matches, skip OpenAI API routing and use chatgpt.com
backend which has primary_assistant mode with web_search tool built in.
Paid accounts still use OpenAI API for non-search queries.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>