Multilingual Support
Benchmarked against: Anthropic — Multilingual support Target: EN / 繁體中文 / 简体中文 Status: Partially implemented — operational multilingual, docs i18n planned
SuperPortia operates in a multilingual environment. The Captain speaks Traditional Chinese, UB entries are in English, and agents communicate in both. This page defines the language strategy.
Language zones
| Zone | Language | Why |
|---|---|---|
| UB entries | English | Cross-agent consistency, better embedding quality |
| UB tags | English | Controlled vocabulary, machine-readable |
| Captain communication | Traditional Chinese | Captain's preferred language |
| Agent-to-agent messages | English or Chinese | Depends on context |
| Docs site | English (primary), 繁中/简中 (planned) | International structure, local accessibility |
| Commit messages | English | Git convention |
| Code comments | English | Developer convention |
| WO titles | English | Searchable, consistent |
| WO descriptions | Mixed | English preferred, Chinese acceptable |
UB language policy
Captain decision (2026-02-28): All UB entries must be in English.
| Content | Language | Exception |
|---|---|---|
| Entry title | English | None |
| Entry content | English | Chinese quotes from source material allowed |
| Entry tags | English | None — controlled vocabulary |
| Entry entities | English | Proper nouns may include Chinese |
| Search queries | English recommended | Chinese works for semantic (Gemini embedding supports it) |
Why English?
- Cross-agent consistency — All agents read the same language
- Better embedding quality — Gemini
embedding-001handles English well - Controlled Vocabulary — CV is in English
- International readability — Future team members, open documentation
Embedding model language support
| Model | English | Chinese | Notes |
|---|---|---|---|
Gemini embedding-001 (768d) | Excellent | Good | Cloud UB uses this |
| Previous: mem0 + Qdrant | Good | Poor | Local UB had Chinese issues |
Historical issue (2026-03-02): Local UB (mem0 + Qdrant) couldn't do Chinese semantic search. Searching Chinese keywords returned no results even when entries existed. This was a key driver for the Cloud UB unification.
Agent communication language
| Scenario | Language | Example |
|---|---|---|
| Agent → Captain | Traditional Chinese | Status reports, proposals |
| Captain → Agent | Traditional Chinese | Instructions, feedback |
| Agent → Agent (mailbox) | English or Chinese | Depends on content |
| Agent → UB (ingestion) | English | All entries |
| Agent → Code (commits) | English | Git messages |
| Agent → Docs | English | Documentation pages |
Docs site i18n (planned)
Docusaurus has built-in i18n support. Target structure:
docs-site/
├── docs/ # English (default)
├── i18n/
│ ├── zh-Hant/ # 繁體中文
│ └── zh-Hans/ # 简体中文
| Phase | Scope | Status |
|---|---|---|
| Phase 1 | English-only docs | Current |
| Phase 2 | 繁體中文 translation | Planned |
| Phase 3 | 简体中文 translation | Planned |
Engine language capabilities
| Engine | English | Chinese | Best for |
|---|---|---|---|
| Claude (all) | Excellent | Excellent | Any language task |
| Groq (Llama 3.3) | Good | Limited | English research |
| Gemini | Good | Good | General multilingual |
| DeepSeek | Good | Excellent | Chinese analysis |
| Zhipu (GLM-5) | Good | Excellent | Chinese NLP, best Chinese support |
| Mistral | Good | Limited | European languages |
For Chinese-heavy tasks, prefer Zhipu or DeepSeek over Groq/Mistral.
Related pages
| Page | Relationship |
|---|---|
| UB Governance | English-only ingestion rule |
| Engine Overview | Engine language capabilities |
| Embeddings | Embedding model details |