Glossary

Glossary is how CrossMeet keeps meeting-grade precision under your control. This guide covers creation, grouping, activation, and best practices.

1. The problem it solves

By default an LLM translates using "general corpus" — your brand names, client names, product codes, and industry jargon can be transliterated or dropped. For example:

• The brand Acme becomes "Pinnacle"
• The product Phoenix-7 becomes "Phoenix Seven"
• The client Schwartz GmbH becomes "Schwartz Limited Liability Company"

Glossary forces the LLM to use your preferred rendering, and verifies the output afterward.

2. Three types

CrossMeet splits terms into three categories, each with its own strategy:

3. Create a term

1. Open CrossMeet → Settings → Glossary
2. Click New Term
3. Fill in:
   • Source (e.g. Acme)
   • Target (translation, e.g. Acme, 量子加密)
   • Type (Brand / Person / Term)
   • Group (which group it belongs to — see below)
   • Notes (optional)
4. Save

4. Groups

Organize terms by meeting type / client / project. Examples:

• acme-2026-q3 — terms for Q3 meetings with Acme
• internal — internal stand-up jargon
• tech-stack — your team's tech vocabulary

When you activate a group, terms from other groups are not injected — preventing cross-contamination.

Create a group

1. Glossary page → Groups tab → New Group
2. Name + description
3. Drag existing terms into it, or pick this group when creating new terms

Activate a group

Before a meeting:

1. Main view → Active Glossary dropdown
2. Check one or more groups
3. Takes effect immediately — new sentences use the new terms

5. How LLM injection works

Before every LLM call, CrossMeet injects the active groups' terms into the prompt:

Translate following text from EN to ZH.
Glossary (MUST follow):
- Acme → Acme
- Phoenix-7 → Phoenix-7
- quantum encryption → 量子加密
...

Text: Acme just rolled out Phoenix-7 with quantum encryption.
Translation: Acme 刚刚发布了 Phoenix-7，搭载量子加密。

After generation, CrossMeet verifies the output: scans whether the expected terms appear, warns or auto-fixes if not.

6. Best practices

• One group per meeting: keep it lean — huge term sets blow up the prompt
• Keep brand names in English unless your client explicitly asks for transliteration
• Prefer keeping person names: English names usually stay English
• 5 min pre-call glossary check: scan for missing terms
• Post-meeting review: add any mistranslated terms for next time

7. Bulk import

CSV import is supported:

source,target,type,group,notes
Acme,Acme,brand,acme-2026,Main client brand
Phoenix-7,Phoenix-7,brand,acme-2026,Core product
quantum encryption,量子加密,term,tech-stack,Crypto term

Settings → Glossary → Import CSV → pick file.

8. FAQ

Q: The LLM still mistranslates a term — what now?

• Check the term is in an active group
• Check the Source field is case-exact (default is case-sensitive)
• Add disambiguating context in Notes (e.g. "tech context only")

Q: Are terms per language pair?

• Yes. Glossaries are stored "source → target" per language pair. The same Source for EN→ZH and EN→JP are two separate rows.

Q: Maximum number of terms?

• No global cap on Pro. But per-call we recommend ≤ 200 active terms (long prompts hurt latency).

Q: Can I import from xlsx?

• Convert to CSV first. CSV is natively supported.

9. Related

• Knowledge base (RAG) — inject document-level context (complementary)
• Quick phrases — one-tap full canned sentences