Persistent, decay-aware memory for Paperclip agents — powered by the Smara Memory API.
Every Paperclip agent gets persistent memory across sessions. Store facts, search by meaning, and recall context — with Ebbinghaus forgetting curves that make recent memories rank higher than stale ones.
# From the Paperclip UI
POST /api/plugins/install
{ "packageName": "@smara/paperclip-plugin-memory" }
# Or local development
POST /api/plugins/install
{ "packageName": "/path/to/paperclip-plugin", "isLocalPath": true }In the Paperclip plugin settings:
| Setting | Required | Description |
|---|---|---|
smaraApiKeyRef |
Yes | Secret reference for your Smara API key |
smaraApiUrl |
No | API URL (default: https://api.smara.io) |
defaultUserId |
No | Default user ID when agents don't specify one |
Get a free API key at smara.io — 10,000 memories, no credit card.
| Tool | Description |
|---|---|
memory-store |
Store a fact with importance scoring (0-1) |
memory-search |
Semantic search with decay-aware ranking |
memory-context |
Pre-formatted context for agent prompts |
memory-delete |
Delete a specific memory |
memory-usage |
Check plan limits and memory count |
score = similarity × 0.7 + decay_score × 0.3
- Similarity: Voyage AI vector embeddings, cosine similarity
- Decay: Ebbinghaus forgetting curve —
e^(-days / (importance × 10)) - Recent + frequently accessed = stronger signal
- Contradictions auto-detected and old facts soft-deleted
MIT