Novyx Memory
The only memory skill with rollback, cryptographic audit trails, and governed actions. 60+ commands covering 100% of Novyx Core — memory, drafts, knowledge graph, context spaces, time-travel replay, eval scoring, execution tracing, and policy-controlled actions.
Why Novyx Memory
Every other memory skill gives you save and search. Novyx gives you the full stack:
| Capability | Novyx | Built-in | SuperMemory | Mem0 | memory-tools |
|---|
| Semantic search | ✓ | ✗ | ✓ | ✓ | ✓ |
| Time-travel rollback |
✓ | ✗ | ✗ | ✗ | ✗ |
| Cryptographic audit trail | ✓ | ✗ | ✗ | ✗ | ✗ |
| Draft review workflow | ✓ | ✗ | ✗ | ✗ | ✗ |
| Knowledge graph | ✓ | ✗ | ✗ | ✗ | ✗ |
| Context spaces (multi-agent) | ✓ | ✗ | ✗ | ✗ | ✗ |
| Governed actions | ✓ | ✗ | ✗ | ✗ | ✗ |
| Execution tracing | ✓ | ✗ | ✗ | ✗ | ✗ |
| Memory health eval | ✓ | ✗ | ✗ | ✗ | ✗ |
| Confidence & decay | ✓ | ✗ | ✗ | ✗ | ✓ |
When to Use This Skill
- - Agent needs memory that persists across sessions
- You need to undo, roll back, or audit what the agent remembered
- Multiple agents need shared context spaces
- Agent actions need policy approval before execution
- You want a knowledge graph built from agent observations
- You need time-travel debugging ("what did the agent know at 3pm?")
- CI pipeline needs a memory health gate
Automatic Behavior
When active, this skill:
- - Auto-recalls relevant memories before each response and injects them as context
- Auto-saves each user message and agent response to persistent storage
- Skips trivial messages (under 15 characters) to conserve API calls
- Filters out previously injected context to prevent feedback loops
Commands (60+)
Memory Basics
| Command | What it does |
|---|
| INLINECODE0 | Save a specific fact or note with optional tags |
| INLINECODE1 |
Semantic search with relevance scores |
|
!list [--tag tag] [--limit N] | List stored memories with optional filters |
|
!forget <topic> | Find and delete all memories matching a topic |
|
!undo [N] | Delete the last N saved memories (default: 1) |
|
!supersede <id> <new text> | Replace an old memory with a corrected version |
Draft Review Workflow
Stage memory changes before committing — like git for agent memory.
| Command | What it does |
|---|
| INLINECODE6 | Create a reviewable draft |
| INLINECODE7 |
List pending drafts |
|
!draft-diff <id> | Show field-level diff for a draft |
|
!merge-draft <id> | Commit draft to canonical memory |
|
!reject-draft <id> | Reject a draft without saving |
|
!branch <name> | View all drafts in a branch |
|
!merge-branch <name> | Merge all drafts in a branch at once |
|
!reject-branch <name> | Reject all drafts in a branch |
Memory Links
| Command | What it does |
|---|
| INLINECODE14 | Create a directed link between memories |
| INLINECODE15 |
Remove a link |
|
!links <memory_id> | Show all links for a memory |
Knowledge Graph
Build structured knowledge from unstructured observations.
| Command | What it does |
|---|
| INLINECODE17 | Add a knowledge triple |
| INLINECODE18 |
Query triples with filters |
|
!entities [--type t] | List all entities |
|
!entity <id> | Get entity with all connected triples |
|
!del-triple <id> | Delete a triple |
|
!del-entity <id> | Delete entity and cascade triples |
|
!graph [--relation r] | View memory graph edges |
Context Spaces (Multi-Agent Collaboration)
Isolated memory namespaces that can be shared between agents.
| Command | What it does |
|---|
| INLINECODE24 | Create a context space |
| INLINECODE25 |
List accessible spaces |
|
!space-memories <id> [query] | List or search within a space |
|
!update-space <id> [--name n] | Update space metadata |
|
!del-space <id> | Delete a space (owner only) |
|
!share <space_id> <email> | Share a space with another user |
|
!shared | List spaces shared with you |
|
!accept <token> | Accept a share invitation |
|
!revoke <token> | Revoke a share |
Replay (Time-Travel Debugging)
Scrub through the full history of your agent's memory.
| Command | What it does |
|---|
| INLINECODE33 | Full timeline of memory operations |
| INLINECODE34 |
Reconstruct memory state at a point in time |
|
!lifecycle <id> | Full biography of a single memory |
|
!replay <id> | Complete history of a memory |
|
!replay-diff <t1> <t2> | Diff memory state between two timestamps |
|
!recall-at <time> <query> | What would search have returned in the past? |
|
!replay-drift <t1> <t2> | Memory composition drift analysis |
Audit & Rollback
| Command | What it does |
|---|
| INLINECODE40 | Last N operations with SHA-256 hashes (default: 10) |
| INLINECODE41 |
Verify cryptographic audit chain integrity |
|
!rollback <time> | Rewind memory to a point in time ("1h", "2 days ago", ISO) |
|
!rollback-preview <time> | Preview what rollback would do without executing |
|
!rollback-history | List past rollback operations |
Cortex (Autonomous Intelligence)
AI that watches your memory and generates insights automatically.
| Command | What it does |
|---|
| INLINECODE45 | Cortex status and last run |
| INLINECODE46 |
Manually trigger a Cortex cycle |
|
!insights | List auto-generated insights |
|
!cortex-config | View current configuration |
Eval (Memory Health Scoring)
Quantify your agent's memory quality with weighted scoring.
| Command | What it does |
|---|
| INLINECODE49 | Run memory health evaluation (recall 40%, drift 30%, conflict 15%, staleness 15%) |
| INLINECODE50 |
CI gate — pass/fail on memory quality (default: 0.7) |
|
!eval-history | Past evaluation scores with trends |
|
!eval-drift | Detect memory drift over time |
|
!health | Quick memory stats overview |
Tracing (Execution Audit)
Track multi-step agent workflows with cryptographic verification.
| Command | What it does |
|---|
| INLINECODE54 | Create a new execution trace |
| INLINECODE55 |
Add a step |
|
!trace-complete <id> | Finalize trace with RSA signature |
|
!trace-verify <id> | Verify trace chain integrity |
Control (Governed Actions)
Policy-evaluated agent actions with approval workflows.
| Command | What it does |
|---|
| INLINECODE58 | List pending approval requests |
| INLINECODE59 |
Approve a pending action |
|
!policy | View active security policies |
|
!actions [N] | Recent action history |
Overview
| Command | What it does |
|---|
| INLINECODE62 | Memory usage, tier, and rollback count |
| INLINECODE63 |
Full dashboard with health score and activity |
|
!context | Current memory context snapshot |
|
!stats | Detailed memory statistics |
|
!help | List all 60+ commands |
Key Differentiators
- - Rollback:
!rollback 1h restores memory to exactly one hour ago. No other memory skill supports this. - Audit trail: Every operation logged with SHA-256 hashes forming a tamper-proof chain.
!audit-verify checks integrity. - Draft workflow: Stage memory changes in branches, review diffs, merge or reject — like git for memory.
- Knowledge graph: Build structured triples from unstructured observations. Query relationships, not just content.
- Context spaces: Isolated memory namespaces with sharing. Perfect for multi-agent architectures.
- Governed actions: Policy-evaluated agent actions with Solo/Team/Enterprise approval workflows.
- Time-travel:
!recall-at 3pm "project status" shows what the agent would have found at 3pm yesterday. - Eval scoring: Quantified memory health with CI gate integration. Catch memory rot before it matters.
- Execution tracing: Cryptographically signed audit trail for multi-step agent workflows.
- Graceful degradation: If rate limits are hit or the API is unavailable, the agent continues working — it never crashes.
Tiers
| Tier | Price | Memories | Rollbacks | Features |
|---|
| Free | $0 | 5,000 | 10/month | Memory, search, audit, rollback |
| Starter |
$12/mo | 50,000 | 100/month | + Drafts, eval, conflict resolution |
| Pro | $39/mo | 500,000 | Unlimited | + Cortex, replay, knowledge graph, tracing |
| Enterprise | $199/mo | Unlimited | Unlimited | + Governed actions, team sharing, SSO |
Configuration
Requires NOVYX_API_KEY environment variable. Get a free key at https://novyxlabs.com (5,000 memories, no credit card).
Options:
- -
autoSave (default: true) — Automatically save messages - INLINECODE72 (default: true) — Automatically recall context before each response
- INLINECODE73 (default: 5) — Max memories to recall per query
