aether

name	aether
description	Connect to Aether to recall memories, save new facts, manage user profile, and search a knowledge graph. Use when users reference past conversations, personal preferences, or want to store/retrieve information.
metadata	{"author":"neo","version":"2.0"}

Aether Skill

You have access to Aether, a personal knowledge system that stores long-term memories, user facts, preferences, and a semantic knowledge graph.

Connection

MCP Server (SSE): {{MCP_URL}} REST API: {{BASE_URL}}/api/v1/ API Key: {{API_KEY}}

For MCP: Connect via SSE transport with Authorization: Bearer {{API_KEY}} For REST: Use Authorization: Bearer {{API_KEY}} header on all requests.

MCP Tools

recall_memories

Call this at the start of EVERY conversation turn. Searches all memories and knowledge chunks for context. Response automatically includes core memory.

Input: { "query": "user's message or topic", "top_k": 10 }
Output: Core memory + list of matching memories with categories

save_memory

Save new facts when you learn something about the user. Automatically deduplicates.

Input: {
  "content": "Concise fact to remember",
  "category": "user_fact | preference | personality | guide | episodic",
  "importance": 1-10
}

Categories:

• user_fact (8-10): Name, location, job, key life facts
• preference (5-7): Communication style, formatting, tone preferences
• personality (5-7): How the AI should behave based on feedback
• guide (3-6): How to use a tool or procedure
• episodic (1-4): Notable conversation events

update_core_memory

Update the permanent user profile. Values merge with existing data.

Input: {
  "key": "user_profile | preferences | ai_personality | active_context",
  "value": { "name": "Neo", "occupation": "developer" }
}

delete_memory

Remove a specific memory by its ID: { "memory_id": "uuid" }

add_knowledge

Process a document or large text into the knowledge graph. For documents, NOT quick facts.

Input: { "text": "long content...", "title": "optional source name" }

search_knowledge_graph

Search graph nodes by semantic similarity: { "query": "search term" }

get_memory_stats

Returns node count, edge count, memory count by category, chunk count.

get_core_memory

Read the full core memory object (user profile, preferences, AI personality, active context). Use at conversation start if not using recall_memories.

REST API

Upload Text

POST {{BASE_URL}}/api/v1/upload
Authorization: Bearer {{API_KEY}}
Content-Type: application/json

{ "text": "content to add", "title": "optional title" }

Upload File

POST {{BASE_URL}}/api/v1/upload-file
Authorization: Bearer {{API_KEY}}
Content-Type: multipart/form-data

file: <binary>

Read Core Memory

GET {{BASE_URL}}/api/core-memory
Authorization: Bearer {{API_KEY}}

Workflow

1. Every turn: Call recall_memories with the user's message — returns core memory + relevant long-term memories in one call
2. When learning new info: Call save_memory with the appropriate category (automatic deduplication)
3. For critical facts (name, preferences): Also call update_core_memory
4. For documents: Use add_knowledge or POST /api/v1/upload
5. Never assume you already know something — always recall first

Resource: Core Memory

The memory://core resource contains the always-present user profile. It is automatically included in every recall_memories response. You can also read it directly via get_core_memory or GET /api/core-memory.