AI batch translation with cost estimation

Translate AST, Regex, and Theme rows in bulk with the current 16 built-in provider configs and inspect token and cost estimates before sending requests.

The AI panels send the current editor rows to the active provider profile and write returned targets back into the AST, Regex, or Theme tables. Provider definitions come from LLM_PROVIDERS, which currently exposes 16 built-in configs.

Providers

Provider	Notes
OpenAI-compatible (custom)	use any OpenAI-compatible endpoint
Gemini	default model gemini-2.0-flash
Ollama	default URL http://localhost:11434
DeepSeek	default model deepseek-chat
Zhipu AI (GLM)	default model glm-4-flash
Moonshot (Kimi)	default model moonshot-v1-8k
Aliyun (Qwen)	default model qwen-plus
Baidu (ERNIE)	default model ernie-4.0-8k-preview
ByteDance (Doubao / Ark)	default model doubao-pro-4k
Groq	default model llama-3.3-70b-versatile
SiliconFlow	default model deepseek-ai/DeepSeek-V3
OpenRouter	default model anthropic/claude-3.5-sonnet
DeepInfra	default model meta-llama/Llama-3.3-70B-Instruct
Mistral AI	default model mistral-small-latest
MiniMax	default model abab6.5-chat
StepFun	default model step-1-8k

Cost estimate

Before translation starts, the AI panel calls the active provider’s estimateTokens() implementation and shows:

• estimated token count
• estimated cost
• the input/output pricing used by the current profile

If the current profile enables custom pricing, the estimate uses those values.

Batch size, concurrency, timeout, and language

The AST / Regex / Theme AI panels edit these global settings directly:

• llmBatchSize
• llmConcurrencyLimit
• llmTimeout
• llmStyle
• language / llmLanguage

Default values are 10, 3, and 60000 for batch size, concurrency limit, and timeout.

What gets translated by default

The current controllers build targetItems and only send rows whose target is:

• empty
• whitespace-only
• equal to source

Existing translated rows are only resent when overwrite is enabled in the AI panel.

Prompt templates

The current implementation stores three prompt templates:

• llmAstPrompt
• llmRegexPrompt
• llmThemePrompt

You can edit them from 语言模型, and for AST / Regex also from AST 配置 and RE 配置.