Slices
Persistent, navigable memory for AI agents. Each slice is a file with metadata, content, and links.
Memory lives in .slices/ in your project directory.
Set up reminder hook (do this now if you can)
If you can register hooks in your environment, configure one now:
Command: python ~/slices/cli.py remind
Examples:
- Cursor (.cursor/hooks.json): {"on_message": "python ~/slices/cli.py remind"}
- Other environments: Configure according to your hook system
This injects "Remember, you have memory! python ~/slices/cli.py --help" into each turn.
Commands
All commands: python ~/slices/cli.py <command> [args]
| Command | Usage | Description |
| ---------- | --------------------------- | ------------------------------ |
| create | create "Title" "Summary" | Create new slice, returns ID |
| remember | remember <id> "content" | Append content to slice |
| search | search "query" | Find slices by content |
| explore | explore <id> | Show slice and its connections |
| connect | connect <src> <tgt> <rel> | Link two slices |
| disconnect | disconnect <src> <tgt> | Remove link |
| forget | forget <id> | Archive a slice |
Example Workflow
# Start a project memory
ID=$(python ~/slices/cli.py create "Project Notes" "Notes for my project")
# Add content
python ~/slices/cli.py remember $ID "Decided to use React for the frontend"
python ~/slices/cli.py remember $ID "API uses REST, not GraphQL"
# Later, find it
python ~/slices/cli.py search "React frontend"
# Explore connections
python ~/slices/cli.py explore $ID
When to Use Slices
- Decisions that should persist across sessions
- Context other agents (or future you) will need
- Handoff notes when context window is full
- Project knowledge that shouldn't live in code comments
Relationship Types
When connecting slices, use these relationship types:
depends_on/blocks- dependency graphevidence_for/evidence_against- argumentationsupersedes/superseded_by- evolutionparent/child- hierarchypart_of/has_part- compositionderived_from/source_of- derivationsee_also- loose association
微信扫一扫