Glossary & Translation Style¶
OLAV's docs are bilingual. When a term has more than one plausible Chinese translation, this table picks the canonical form. Drift — e.g. translating "agent" sometimes as 代理 and sometimes as 智能体 within the same page — makes search broken and confuses readers switching between EN and ZH versions.
If a new term doesn't fit anything below, add it here (in both languages) in the same PR as the first doc that uses it.
Core platform vocabulary¶
| English | Chinese | Notes |
|---|---|---|
| Agent | 代理 | NOT 智能体. OLAV agents are tool-orchestrating LLM workers; the literal "代理" (proxy) is closer to LangChain's framing. "智能体" is the AI-industry buzzword but doesn't match OLAV's actual graph-based execution model. |
| Sub-agent | 子代理 | Compound from "agent". |
| Orchestrator | 编排器 | The top-level agent that routes between sub-agents. Not 调度器 (scheduler — implies cron-style timing). |
| Skill | 技能 | The SKILL.md + tools package an agent loads. Not 能力 (capability — too abstract). |
| Tool | 工具 | LangChain @tool-decorated function. Uniform across docs. |
| Workspace | 工作空间 | .olav/workspace/<agent>/ runtime tree. Not 工作区. |
| Memory (semantic) | 记忆 | LanceDB-backed retrieval store. |
| Memory (RAM) | 内存 | Only when literally referring to process RAM / DuckDB in-memory clones. Don't mix. |
| Knowledge Base / KB | 知识库 | The user-facing name for the UKS memory table contents. |
| Guide | guide | Keep the English word inline (e.g. *.guide.yaml). Use 指南 only in headings of long-form tutorials, never as a translation of the YAML file type. |
| Tier (source-tier) | 分层 / tier | "Source-tier" → 来源分层; per-tier names (vendor / platform / team / user) stay English. |
| Reasoning / Thinking | 推理 | Not 思考 (too colloquial). e.g. "thinking mode" → "推理模式". |
| Audit chain | 审计链 | kb_audit/ + audit.duckdb together. |
| Tombstone | tombstone | Loanword; explain on first use per page. |
| Idempotent | 幂等 | Standard. |
| Foreign key (FK) | 外键 / FK | Use English abbreviation in code-adjacent prose. |
| Snapshot | 快照 | Standard. |
| Drift | 漂移 | Standard. |
| Hook | hook | Loanword; in long-form prose may write "勾子". |
| Middleware | middleware | Loanword. Don't translate. |
| Callback | callback | Loanword. |
R102 / R88 / R85 references¶
These appear throughout dev_docs and changelogs:
| Token | Means |
|---|---|
| R102 | Conversational memory ingestion (the memory_curator sub-agent path). |
| R88-A | Per-service tool generation at registration time (ADR-0007 precursor). |
| R85 | Inline-save / format_and_export write-class tool. |
Keep these as-is in both languages.
Capitalisation in code-adjacent prose¶
- Agent / sub-agent / tool names stay lowercase as in source:
core,netops,analyzer,sim,memory_curator. - SKILL.md / AGENT.md / MANIFEST.yaml stay all-caps with
.md/.yaml. - CLI verbs are lowercase as typed:
olav user create,olav kb remove. - Don't translate them in either language.
When you must coin a new term¶
Pick one form, then update this table in the same commit. If the English term is ambiguous in Chinese (like "memory" → 记忆 vs 内存), disambiguate by context first, by convention second.