Quickstart Guide¶
Get a fully functional AI memory system in one command. No 26-step manual setup.
See It In Action¶
3 Steps — That's It¶
1
Install
pip install neural-memoryPython 3.11+, works on Windows, Linux, macOS.
2
Set Up Everything
nmem init --fullConfig, brain, MCP, hooks, embeddings, dedup — all automatic.
3
Restart Your IDE
Restart Claude Code, Cursor, or your MCP client.
Your brain is live. Start talking.
Your brain is live. Start talking.
What You Get¶
46 MCP tools
11 diagnostic checks
3 auto-installed hooks
Semantic search via embeddings
Duplicate detection
Verify Your Setup¶
Run nmem doctor to see if everything is healthy:
Something wrong? Run nmem doctor --fix to auto-remediate hooks, dedup, and embedding issues.
Core Features¶
🧠
Remember
Store memories with type, priority, and tags.
Rich context creates strong neural connections.
nmem_remember("chose Redis over Memcached because...")
🔍
Recall
Retrieve by meaning, not keywords.
Spreading activation finds related memories across your graph.
nmem_recall("caching strategy")
🩺
Doctor
11 diagnostic checks. Auto-fix with
--fix.
Know exactly what's configured and what's missing.
📄
Knowledge Surface
A
.nm file loaded every session — working memory.
Depth-aware routing skips the database when the surface is sufficient.
What --full Does (Under The Hood)¶
| Step | What | Why |
|---|---|---|
| Config | Creates ~/.neuralmemory/config.toml |
Central settings for brain, embedding, dedup |
| Brain | Creates default.db SQLite database |
Your memory graph lives here |
| MCP | Registers in Claude Code + Cursor | AI tools can use your brain |
| Hooks | Installs PreCompact, Stop, PostToolUse | Captures memories automatically |
| Embeddings | Auto-detects best provider | Semantic search (find by meaning) |
| Dedup | Enables duplicate detection | Prevents memory bloat |
| Maintenance | Generates maintenance.sh/.ps1 |
Scheduled decay + consolidation |
Without --full, run nmem init for the basics, then nmem setup embeddings separately.
Power User Features¶
Cognitive Reasoning¶
Build and test hypotheses across sessions:
nmem_hypothesize("Redis is better than Memcached for our read pattern")
nmem_evidence("mem_id", "supports", confidence=0.8)
nmem_predict("Latency will drop 40% after Redis migration", deadline="2026-04-01")
nmem_verify("prediction_id", outcome="correct")
Cloud Sync¶
Sync your brain across devices:
Background Maintenance¶
Keep your brain healthy automatically:
Or use the generated maintenance script:
# Linux/macOS — add to crontab
crontab -e
0 3 * * * ~/.neuralmemory/maintenance.sh
# Windows — run via Task Scheduler
~\.neuralmemory\maintenance.ps1
Need Help?¶
- Issues? Run
nmem doctor --fixfirst - Questions? GitHub Discussions
- Bugs? Open an issue
- Full docs? You're already here — check the sidebar