memq-memory

name	memq-memory
description	Hosted governed memory fabric for AI coding agents via MemQ MCP tools. Provides session lifecycle hooks, memory patterns, and best-practice recipes for durable recall, episodic replay, and cross-session continuity.
version	1.0.0
homepage	https://multinex.ai/memq
tools	["add_memory","query_memory","search_memory","recent_memory","reflect_memory","save_context","memory_status","health_check","reunion","mnemosyne_context"]

MemQ Memory Skill

Hosted governed memory fabric for AI coding agents. This skill teaches agents how to use MemQ MCP tools effectively for durable recall, session continuity, and cross-conversation intelligence.

Prerequisites

MemQ MCP is hosted at https://mcp.multinex.ai/mcp/v1 and governed through Billing Manager OAuth/API-key access. Start hosted access through the Billing Manager: MemQ signup.

Install via:

curl -fsSL https://memq.multinex.ai/install-mcp | bash

Or manually add to your MCP client configuration:

{
  "mcpServers": {
    "memq": {
      "url": "https://mcp.multinex.ai/mcp/v1",
      "auth": "oauth2",
      "discoveryUrl": "https://billing.multinex.ai/.well-known/oauth-authorization-server"
    }
  }
}

Session Lifecycle Hooks

On Session Start

Execute these steps immediately when a new conversation begins:

1. reunion() → Handshake with MemQ, get namespace status and protocol version
2. recent_memory(limit: 10) → Load last 10 memories for immediate context
3. search_memory(query: "<current task description>", top_k: 5) → Find relevant past work

Example:

→ reunion(include_status: true)
← { protocol: "2.7", namespace: "org/project", status: "healthy" }

→ recent_memory(limit: 10)
← [{ text: "Fixed auth bug in middleware...", type: "checkpoint" }, ...]

→ search_memory(query: "billing checkout flow", top_k: 5)
← [{ text: "Stripe webhook handler needs...", score: 0.92 }, ...]

During Work

Use these patterns while working on tasks:

When	Tool	Example
Before planning/decisions	search_memory	search_memory(query: "similar authentication pattern", top_k: 5)
After completing a task	add_memory	add_memory(text: "Implemented retry logic for API calls using exponential backoff", memory_type: "checkpoint", tags: ["api", "retry"])
On encountering errors	add_memory	add_memory(text: "TypeError in auth middleware caused by null session — fixed by adding guard clause", memory_type: "episodic", tags: ["bug", "auth"])
On discovering patterns	add_memory	add_memory(text: "This codebase uses Zod for all IO boundary validation", memory_type: "semantic", tags: ["pattern", "validation"])
Saving important context	save_context	save_context(content: "Architecture decision: chose Workers over Lambda for edge latency", kind: "semantic")

On Session End

Consolidate learnings before the conversation closes:

reflect_memory(window: 50, force: true) → Compress and consolidate recent work into durable patterns

Memory Types

Type	When to Use	Example
episodic	Events, errors, debugging sessions	"Fixed race condition in checkout flow"
semantic	Facts, patterns, architecture knowledge	"Auth uses HMAC-SHA256 with 30s replay window"
procedural	How-to knowledge, workflows	"Deploy via: push to master → CI → wrangler deploy"
checkpoint	Task completion summaries	"Completed billing dashboard overhaul: 7 files modified"
reflection	Meta-learning, strategy insights	"Complex auth bugs are faster to fix by checking middleware first"

Advanced Patterns

Context Packing

Use mnemosyne_context to get a pre-packed context bundle from all memory tiers:

mnemosyne_context(objective: "Debug the failing webhook handler", include_recent: true, include_learning: true)

Returns a HOT/WARM/COLD/ENGINE context pack optimized for the given objective.

Episodic Replay

When debugging similar issues to past ones, use brain tools:

brain_recall_episode(cue: "webhook timeout failure", include_related: true, limit: 10)
brain_associate(cue: "stripe checkout error", include_temporal: true)

Predictive Correction

Before executing risky operations, check predictions:

brain_predict(objective: "Deploying new auth middleware to production", limit: 10)

Returns forecasted risks, guardrails, and corrective patterns from past experience.

Sleep Consolidation

For long-running projects, run periodic consolidation:

brain_consolidate_sleep(limit: 100, persist: true, include_commons_candidates: true)

Compresses context, extracts semantic rules, and surfaces patterns worthy of shared commons promotion.

Tagging Best Practices

Use consistent, hierarchical tags for better retrieval:

Category	Tags
Domain	auth, billing, api, ui, database
Action	bug, feature, refactor, deploy, debug
Severity	critical, minor, optimization
Project	memq, billing-manager, nexus-ranger

Health Monitoring

Check MemQ system health at any time:

health_check() → Aggregate backend status
memory_status() → Detailed telemetry, limits, and tier config
namespace_info() → Authenticated namespace boundary and billing posture

Degraded Mode

If MemQ MCP tools are unavailable:

1. Emit warning: "⚠️ MemQ unavailable — operating in degraded mode (no persistent memory)"
2. Continue execution without memory operations
3. This is acceptable for offline development