Agent Setup
Add Nella instructions to your project so your agent uses it on every task — no manual prompting needed.
Add Nella instructions to your project so your agent uses it on every task — no manual prompting needed.
Always-On Nella
Most AI coding agents read project instruction files at the start of each conversation. Add Nella usage patterns to these files and your agent will search, track assumptions, and verify context automatically.
Claude Desktop & Claude Code
Add to your CLAUDE.md or project-level Claude.md:
## Nella
Before making changes, search the codebase with `nella_search` to understand existing patterns.
Record assumptions with `nella_add_assumption` before implementing.
After changes, verify with `nella_check_assumptions` and `nella_check_dependencies`.
Cursor
Add to .cursorrules in your project root:
## Nella
Always search the codebase with nella_search before generating code.
Record assumptions about the codebase with nella_add_assumption.
After making changes, run nella_check_assumptions to verify nothing broke.
VS Code / GitHub Copilot
Add to .github/copilot-instructions.md:
## Nella
Use nella_search to find existing implementations before writing new code.
Track assumptions with nella_add_assumption for later verification.
Run nella_check_dependencies after modifying package.json.
Windsurf
Add to .windsurfrules:
## Nella
Search before coding: use nella_search to find relevant existing code.
Track what you assume: use nella_add_assumption to record beliefs.
Verify after changes: use nella_check_assumptions to catch contradictions.
Prompting Tips
Good patterns:
- “Search for the existing implementation before writing new code”
- “Record your assumption about the database schema”
- “Check assumptions after modifying the auth module”
- “Verify dependencies haven’t changed”
Patterns to avoid:
- “Always use all Nella tools” — too vague, leads to unnecessary calls
- “Search for everything” — too broad, produces noise
- “Index before every search” — unnecessary, the index persists
Tip
Start simple. A few lines in your project instructions is enough. You can refine based on how your agent behaves.