Neutron Agent Memory Skill

Persistent memory storage with semantic search for AI agents. Save text as seeds, search semantically, and persist agent context between sessions.

Prerequisites

API credentials via environment variables:

export NEUTRON_API_KEY=your_key
export NEUTRON_APP_ID=your_app_id
export NEUTRON_EXTERNAL_USER_ID=1  # optional, defaults to 1

Or stored in ~/.config/neutron/credentials.json:

{
  "api_key": "your_key_here",
  "app_id": "your_app_id_here",
  "external_user_id": "1"
}

Testing

Verify your setup:

./scripts/neutron-memory.sh test  # Test API connection

Scripts

Use the provided bash script in the scripts/ directory:

• neutron-memory.sh - Main CLI tool

Common Operations

Save Text as a Seed

./scripts/neutron-memory.sh save "Content to remember" "Title of this memory"

Semantic Search

./scripts/neutron-memory.sh search "what do I know about blockchain" 10 0.5

Create Agent Context

./scripts/neutron-memory.sh context-create "my-agent" "episodic" '{"key":"value"}'

List Agent Contexts

./scripts/neutron-memory.sh context-list "my-agent"

Get Specific Context

./scripts/neutron-memory.sh context-get abc-123

Interaction Seeds (Dual Storage)

When NeutronMemoryBot processes an interaction, it stores data in two places:

• Agent Context - Truncated summary for structured metadata and session tracking
• Seed - Full thread snapshot for semantic search

• Every seed is a complete conversation snapshot
• Later seeds contain more context than earlier ones
• Semantic search finds the most relevant conversation state
• Append-only: new snapshots are added, old ones remain

Seed Format

Thread snapshot - {timestamp}

Post: {full post content}

Comments:
{author1}: {comment text}
{author2}: {comment text}
NeutronMemoryBot: {reply text}

API Endpoints

• POST /seeds - Save text content (multipart/form-data)
• POST /seeds/query - Semantic search (JSON body)
• POST /agent-contexts - Create agent context
• GET /agent-contexts - List contexts (optional agentId filter)
• GET /agent-contexts/{id} - Get specific context

Memory types: episodic, semantic, procedural, working

Text types for seeds: text, markdown, json, csv, claude_chat, gpt_chat, email