Skip to main content

Universal Reference

← Back to Agent Skills

File: AUDAXIC_MEMORY.md (project root)

Any agent can read this file. Use it as a standalone reference or import it into your agent's config.

# Audaxic Memory — Procedural Guide

## Overview

Audaxic provides persistent memory for AI agents via MCP (Model Context Protocol).

- **MCP Server**: `https://memory-mcp.audaxic.com/mcp`
- **Auth**: Bearer token (user's API key)
- **Prerequisite**: MCP server must be connected before using memory tools

## MCP Connection Check

Before using memory tools, verify the MCP server is connected. If you encounter connection errors, instruct the user to test their connection:

npx @modelcontextprotocol/inspector --transport http --server-url https://memory-mcp.audaxic.com/mcp --header "Authorization: Bearer ak_your_key_here"

If the inspector returns a successful response, the MCP server is reachable. If not, ask the user to verify their API key and network connection.

## Behavioral Triggers

### When to AUTO-SEARCH (before acting)

- Before making architectural decisions → search for prior decisions
- Before adding new dependencies → search for existing preferences
- When user mentions a past conversation → search for related context
- Before fixing a bug → search for known issues/workarounds
- When starting a new session → search for user preferences

### When to AUTO-SAVE (after learning)

- After user states a preference → save as memory
- After making a decision with rationale → save decision + reasoning
- After fixing a complex bug → save solution pattern
- After learning project conventions → save conventions
- After user provides feedback → save feedback context

### When NOT to Save

- Temporary state (use variables/cache)
- Ephemeral debug info
- Information already in code comments
- Secrets, API keys, passwords

## MCP Tools

### add_memory

Store a new memory.

Parameters:
- `text` (required): The memory content to store
- `user_id` (optional): Override user ID (defaults to authenticated user)
- `metadata` (optional): Key-value pairs for additional context

Example:
```json
{
  "text": "User prefers dark mode in IDE and uses VS Code",
  "metadata": { "source": "session-2024-01" }
}

search_memories​

Semantic vector search across stored memories.

Parameters:

  • query (required): Natural language search query
  • user_id (optional): Override user ID (defaults to authenticated user)
  • limit (optional): Maximum results (default 10)

Example:

{
  "query": "What UI framework does the user prefer?",
  "limit": 5
}

get_memories​

List all memories for the authenticated user.

Parameters:

  • user_id (optional): Override user ID (defaults to authenticated user)
  • limit (optional): Maximum results (default 100)

delete_memory​

Delete specific memories by their IDs.

Parameters:

  • memory_ids (required): Array of memory ID strings

Example:

{
  "memory_ids": ["mem_abc123", "mem_def456"]
}

delete_all_memories​

Delete all memories for the authenticated user.

Parameters:

  • user_id (optional): Override user ID (defaults to authenticated user)

Content Best Practices​

  • One fact per memory — avoid bundling unrelated information
  • Use natural language — "Chose React over Vue for dashboard due to component library ecosystem" not "React chosen"
  • Include context and reasoning — store decisions, not just outcomes
  • Be specific — better specificity = better semantic matching

Memory Lifecycle​

Add → Search → Update (delete + re-add) → Delete (when obsolete)

**Setup:**

```bash
# Copy to project root
# No directory structure required
Last updated on