1 min lesson
Local vs server-side: the central tradeoff
Respond to "Why does a remote vector index complicate the claim that server-side inference 'keeps code confidential'?" Name the reason and the detail behind it.
Step 1 of 2
Local vs server-side: the central tradeoff
Where computation happens is the real design fork. Server-side gives you latest-generation models and shared indexes; local keeps code on the user's machine but caps capability. The interview wants you to reason about the spectrum, not pick a dogma.
- Concern
- Model power
- Server-side
- Full latest-generation models, big context
- Local / on-prem
- Limited by the user's hardware
- Concern
- Code exposure
- Server-side
- Code leaves the machine (encrypted, not retained)
- Local / on-prem
- Code never leaves; strongest privacy story
- Concern
- Indexing
- Server-side
- Fast shared index, but vectors derived from code live remotely
- Local / on-prem
- Index stays local; slower, heavier on the client
- Concern
- Fit
- Server-side
- Most users; the default product
- Local / on-prem
- Regulated, air-gapped or zero-egress customers
| Concern | Server-side | Local / on-prem |
|---|---|---|
| Model power | Full latest-generation models, big context | Limited by the user's hardware |
| Code exposure | Code leaves the machine (encrypted, not retained) | Code never leaves; strongest privacy story |
| Indexing | Fast shared index, but vectors derived from code live remotely | Index stays local; slower, heavier on the client |
| Fit | Most users; the default product | Regulated, air-gapped or zero-egress customers |
A subtle point that scores well: a remote vector index isn't "not the code." Embeddings can leak structure and can sometimes be partially inverted, so a privacy-sensitive customer cares where the index lives, not just where the raw files live. Treating derived artifacts as sensitive too is the kind of rigor Cursor's truth-seeking culture rewards.