Auto_Bangumi/docs/en/config/llm.md

LLM Parser

The LLM parser calls a large language model to parse RSS titles when regex parsing fails or when higher-quality parsing is desired.

{width=700}{class=ab-shadow-card}

Parse Mode

• Fallback (regex first): default. Use regex first, then LLM if regex fails.
• Primary (LLM first): use LLM first, then regex as a safety net.

Providers

Built-in providers include:

• OpenAI, including OpenAI-compatible endpoints such as DeepSeek, Ollama, LM Studio, OpenRouter and OneAPI.
• Anthropic
• Google Gemini

The settings page can also show installed presets, subscription-account providers and downloadable providers. Subscription providers need to be connected first; downloadable providers may show a risk notice before installation.

LLM requests use the app proxy settings.

config.json

Section: llm

Key	Description	Type	WebUI field	Default
enable	Enable LLM parsing	boolean	Enable LLM Parsing	false
provider	Provider ID	string	Provider	openai
api_key	API key for built-in OpenAI provider	string	API Key	""
model	Model name	string	Model	gpt-5-mini
base_url	OpenAI-compatible endpoint	string	Base URL	""
mode	fallback or primary	string	Parse Mode	fallback
timeout	Request timeout	number seconds	Advanced Settings	20
cache_ttl	Parse cache TTL	integer seconds	Advanced Settings	900
max_concurrency	Maximum concurrent requests	integer	Advanced Settings	2
failure_threshold	Failures before circuit breaker	integer	Advanced Settings	3
failure_backoff	Circuit breaker pause	integer seconds	Advanced Settings	300
providers	Per-provider overrides	object	saved by provider	{}