Semantic code intelligence for AI agents.
40–70× token reduction. Zero data exfiltration. Enterprise-ready.
🆕 v0.9.0 — Enterprise-Ready. docker pull ghcr.io/dfrostar/neuralmind, CycloneDX SBOM on every release, air-gapped install walkthrough, NIST/SOC 2/GDPR one-pager. Release notes → · Air-gapped walkthrough → · Summary →
Earlier: v0.8.0 always-on (systemd/launchd/healthz) · v0.7.0 install anywhere · v0.6.0 live activity feed · v0.4.0 brain-like synapse layer.
~800 tokens per code question instead of 50,000+. Real-world bills drop 40–70%.
Your code never leaves your machine. Zero cloud APIs, zero telemetry.
Full audit trail and enterprise compliance reporting.
Claude Code, Cursor, ChatGPT, Gemini, or any LLM.
Traditional: Load entire files → 50,000+ tokens. NeuralMind: Smart context → ~800 tokens.
A 4-layer semantic index surfaces only the context your question needs—not the entire codebase.
PostToolUse hooks compress Read/Bash/Grep output 88–91% smaller before agents see it.
A persistent weighted graph runs alongside the LLM, learning Hebbian associations between co-active code nodes. Decay prunes stale links; long-term potentiation protects frequently-used ones; spreading activation surfaces related code on every prompt — no MCP call required.
neuralmind serve opens an Obsidian-style force-directed graph of your codebase in the browser. Structural edges, the learned synapse overlay, backlinks, a semantic quick-switcher, and one-click "open in editor". Makes the brain inspectable — no more black-box retrieval.
50K+ tokens of raw source compressed to ~800 tokens of structured context per code question. Cost bills drop 40–70%. The brain-like layer makes retrieval get sharper the longer NeuralMind runs on a codebase.
Every query logged. Export for compliance and auditor review.
ChromaDB, PostgreSQL pgvector, or LanceDB. Choose your infrastructure.
RBAC, rate limiting, secret detection, audit logging.
Full transparency, no vendor lock-in, zero exfiltration.
| Feature | NeuralMind | Cursor @codebase | Claude Projects | Long Context |
|---|---|---|---|---|
| Works everywhere | ✅ Yes | ❌ Cursor only | ⚠️ Claude only | ✅ Yes |
| 100% local & offline | ✅ Yes | ❌ No | ❌ No | ❌ No |
| Token reduction | 40–70× | 2–3× | 0× (loads all) | 0× (loads all) |
| Enterprise compliance | ✅ NIST AI RMF | ⚠️ Basic | ⚠️ Basic | ⚠️ Basic |
# 1. Install NeuralMind
pip install neuralmind
# 2. Install graphify (required)
git clone https://github.com/safishamsi/graphify.git
cd graphify && pip install -e .
# 3. Go to your project
cd /path/to/your-project
# 4. Generate the code graph
graphify update .
# 5. Build the neural index
neuralmind build .
# 6. Test it
neuralmind stats .Reduce monthly AI bills by 40–70% while improving answer quality.
Process sensitive code without exfiltration. GDPR/HIPAA compliant.
Scale AI help to 100K+ LOC without loading entire files.
ChatGPT, Gemini, local models, Claude Code. Works with all of them.
First-time setup for all platforms.
Auto-discovery, cloud sync, CI/CD.
All commands and flags.
NIST AI RMF, audit trails, security.
Versioning, support timeline, upgrades.
Common issues & solutions.
Graph view (neuralmind serve) — Obsidian-style force-directed graph over the index and synapse store, with backlinks, semantic quick-switch, and one-click open-in-editor. Stacks on v0.5.0's bundled MCP server and v0.4.0's brain-like synapse layer.
Graph-view Phase B: replay-last-query overlay (#105), edge tooltips + min-weight synapse slider (#106), pin UX, Cmd-K quick-switch. Then Phase C: live activity feed of synapse co-activations.
Stable API, PostgreSQL pgvector scale, observability dashboard, 2-year LTS.