Force English to keep your AI sharp. Force local language to keep it relevant.
An OpenClaw plugin that pins your AI's output language, preventing it from drifting into the wrong language mid-conversation. Dual injection makes it stick.
Models degrade when they drift away from English mid-conversation. A user writes in Japanese, the AI starts responding in Japanese, and reasoning quality drops. Force English to maintain peak performance.
Perfect for non-English deployments. User writes in English? AI answers in Japanese. User writes in Japanese? AI still answers in Japanese. Pin the output language to match your local context.
Other prompt-level solutions get buried and ignored by the model. Language Anchor injects at two points in every turn:
- System context — persistent system-level override
- Turn context — injected right before each user message
Dual injection makes the instruction much harder to ignore.
30 languages: English, 中文, हिन्दी, Español, العربية, বাংলা, Português, Русский, اردو, Bahasa Indonesia, Deutsch, 日本語, Kiswahili, मराठी, తెలుగు, Türkçe, தமிழ், 한국어, Tiếng Việt, Italiano, Yorùbá, Hausa, Afaan Oromoo, ਪੰਜਾਬੀ, فارسی, Basa Jawa, 吴语, ગુજરાતી, Soomaali, Magyar.
openclaw plugins install https://github.com/lykeion-dev/language-anchorAdd to your openclaw.json:
{
"plugins": {
"language-anchor": {
"enabled": true,
"config": {
"language": "en"
}
}
}
}| Option | Type | Default | Description |
|---|---|---|---|
language |
string |
"en" |
ISO 639-1 language code |
targetAgents |
string[] |
["*"] |
Agent IDs to target (supports * wildcard) |
targetModels |
string[] |
["*"] |
Model name patterns (supports * wildcard) |
targetSessions |
string[] |
["*"] |
Session keys to target (supports * wildcard) |
Force English to keep reasoning sharp:
{
"language": "en"
}Force Japanese for a local deployment:
{
"language": "ja"
}Target only specific models:
{
"language": "en",
"targetModels": ["dashscope/*", "zai/*"]
}MIT