Agent Configuration

To get the most out of Kumbukum with AI coding agents (GitHub Copilot, Cursor, Windsurf, Claude Code, etc.), add an AGENTS.md file to the root of your project. This tells agents how to use Kumbukum as their persistent memory layer and shared knowledge store.

Why?

Without instructions, AI agents don't know Kumbukum exists. An AGENTS.md file tells them to:

• Search before acting — Check Kumbukum for existing context before starting work
• Store learnings — Save decisions, patterns, and outcomes after completing tasks
• Use notes for documentation — Keep project knowledge in Kumbukum, not scattered in files
• Connect knowledge — Link related items so context compounds instead of resetting

Recommended AGENTS.md

Copy the following into an AGENTS.md file at the root of your project:

# Project Instructions

## Knowledge Management
This project uses Kumbukum as its knowledge store via MCP.

### Before Starting Any Task
1. Call `recall_memory` or `search_knowledge` with a query describing the task to check for relevant prior context, decisions, or notes
2. Review any related notes with `search_notes`
3. Use the returned context to inform your approach

### After Completing Significant Work
1. Call `store_memory` to save key decisions, outcomes, and context for future sessions
2. Use descriptive titles and tag memories for easy retrieval
3. Use `create_link` to connect newly created items to related notes, memories, or URLs in the knowledge graph

### Creating Notes
Use `create_note` for structured documentation:
- Architecture decisions
- API designs
- Meeting notes
- Technical specs

After creating a note, use `create_link` to connect it to related items.

### Creating Memories
Use `store_memory` for agent-scoped learnings:
- Debugging insights and solutions
- User preferences and patterns
- Task outcomes and what worked
- Codebase conventions discovered during work

After storing a memory, use `create_link` to connect it to related notes, URLs, or other memories.

### Saving URLs
Use `save_url` to bookmark and extract content from web pages.

After saving a URL, use `create_link` to connect it to related notes or memories.

### Searching
- `search_knowledge` — Search across ALL types (notes, memories, URLs). **Use this first.**
- `search_notes` — Search only notes
- `recall_memory` — Search only memories
- `search_urls` — Search only saved URLs

### Tagging
- Before creating tags, call `suggest_memory_tags` to reuse existing tags and avoid duplicates
- Use consistent, descriptive tags (e.g., `architecture`, `debugging`, `api-design`)

### Knowledge Graph
- Use `create_link` to connect related notes, memories, and URLs
- Use `traverse_graph` to explore connections from a known item
- Use `get_graph` to see the full picture

Customizing

Adapt the template to your workflow. Common additions:

• Project-specific tags — Define standard tags for your domain (e.g., frontend, backend, database)
• Team conventions — Note which types of knowledge go into notes vs memories
• Search-first rules — Require agents to search before creating duplicates

Where to Place It

AI Client	File Location
GitHub Copilot	AGENTS.md in repo root (or any directory)
Cursor	.cursor/rules/ directory
Windsurf	.windsurfrules file
Claude Code	CLAUDE.md in repo root

The content is the same — only the file name and location differ per client.