How does NeuroNest protect against prompt injection attacks?

NeuroNest uses a dual-layer firewall. Layer 1 is a zero-dependency TypeScript engine that applies sanitization, regex-based injection detection, secrets scanning, and policy enforcement in under 1 ms. Layer 2 adds LLM-powered semantic analysis to catch sophisticated attacks that evade pattern matching.

Does NeuroNest send my code to the cloud?

No. NeuroNest is local-first. Source code is processed on your machine by default and is never sent to NeuroNest servers. When you configure a cloud AI provider such as OpenAI or Anthropic, code is sent only to that provider under their privacy policy.

What is the NeuroNest Edit Lock System?

The Edit Lock System lets you mark any file or directory as protected. No agent — including the master orchestrator — can modify locked paths without explicit human approval. This is the seventh security layer in NeuroNest's defense-in-depth model.

What is HMAC request signing in NeuroNest?

All inter-agent communication in NeuroNest is signed with HMAC-SHA256 with nonce-based replay protection. This prevents spoofing of agent messages and replay attacks where a captured valid message is re-submitted.

NeuroNest is an agent-first IDE built on Electron for production development teams. It orchestrates 117+ specialized AI agents across 13 departments with a 9-stage AI pipeline, dual-layer policy engine, live code knowledge graph, local-first inference, and 50+ AI-powered features. No cloud dependency required.

How much does NeuroNest cost?

The Community edition is free forever and includes all 117+ AI agents, swarm orchestration, 11 AI providers, knowledge graph, 7-layer firewall, all channels, skills and templates, recipes, session sharing, and local auth. Professional ($29/month) and Enterprise (custom pricing) plans are coming soon.

What operating systems does NeuroNest support?

NeuroNest currently runs on macOS 13 or later (Apple Silicon and Intel). It works fully offline with local-first inference so your data never leaves your workstation.

What makes NeuroNest different from other AI coding tools?

NeuroNest is purpose-built — not a VS Code fork. It ships with 117+ specialized agents across 13 departments, a 9-stage AI pipeline with ZERA Optimizer, a dual-layer policy engine, a 7-layer defense-in-depth security model, a live code graph, 11 AI providers (local and cloud), and 50+ production-grade features including Recipes, Subagents, OS Mode, P2P Session Sharing, Voice-to-Code, Adversary Reviewer, and more. Data never leaves your workstation.

What are the 13 agent departments in NeuroNest?

Engineering (15 agents), Security (8), Data (10), DevOps (8), Design (6), QA (6), Product (5), Research (5), Mobile (4), Blockchain (4), AI/ML (6), Infrastructure (5), and NeuroNest Orchestration (1). Total: 117+ specialized agents.

What AI providers does NeuroNest support?

NeuroNest supports 11 AI providers. Cloud providers (API key auth): OpenAI, Anthropic, Google Gemini, DeepSeek, Grok (xAI), Mistral, NVIDIA NIM, Groq. Local providers (no auth required): Ollama, llama.cpp, OpenMythos. You can route sensitive operations to local providers so data never leaves your machine.

Is NeuroNest open source?

The NeuroNest repository is available on GitHub at github.com/NETGVai/NeuroNest. The Community edition is free forever.

What are the 5 execution topologies in NeuroNest?

Sequential (agents work in a linear chain — design then implement then test), Star (a hub agent dispatches work to independent spoke agents), Hierarchy (a manager agent coordinates a pool of specialist workers), Mesh (agents communicate peer-to-peer with no central coordinator), and Hybrid (adaptive combination chosen automatically by the orchestrator at runtime).

How does the 7-layer security model work?

Layer 1: Firewall Engine — regex-based sanitization, prompt injection detection (100+ rules, 12 threat categories), secrets scanning, and policy enforcement under 1 ms. Layer 2: Enhanced Firewall — hybrid regex plus LLM semantic analysis for sophisticated attacks, 4 policy presets, PII redaction for 12 data types. Layer 3: Action Analyzer — pre-execution risk scoring for every shell command and file operation using 200+ risk signatures and ensemble worst-case scoring. Layer 4: Runtime Protection — DevTools detection, debugger-flag scanning, SHA-256 file integrity verification, environment anomaly detection, keyboard shortcut blocking. Layer 5: Secure Communication — HTTPS enforcement, TOFU certificate pinning, HMAC-SHA256 request signing, 5-minute replay protection window, macOS Keychain token storage. Layer 6: Native C++ Module — SHA-256, HMAC, AES-256-GCM compiled into a Node-API addon so crypto secrets never surface in JavaScript. Layer 7: Edit Lock — directory-scoped write restrictions so agents can only modify paths they have been explicitly granted access to.

What is the ZERA Optimizer?

ZERA Optimizer is NeuroNest's prompt enhancement layer that runs at Stage 2 of the 9-stage AI pipeline — between user input and the Orchestrator Planner. It improves instruction clarity, reduces ambiguity, and structures user intent before routing to the appropriate agent department.

What are Recipes in NeuroNest?

Recipes are portable YAML workflow configurations with {{variable}} template substitution. They let you save and share multi-step agent workflows. NeuroNest ships 4 built-in recipes: Code Review, API Builder, Security Audit, and Test Generator. Custom recipes can be created and shared via deeplink.

What productivity features does NeuroNest include?

NeuroNest includes a comprehensive feature catalog: Monaco Editor (100+ languages), Inline Code Completion, Embedded Browser with pinned tabs, Extension System (CSV/Markdown/JSON/Image viewers), Multi-Session Parallel Agents, Subagents, Recipes, OS Mode (screen capture), Voice-to-Code (Whisper), P2P Session Sharing (PIN-protected), Global Search, Git Integration (AI commit messages and worktrees), Context Loading (images/URLs/sticky notes), Decision Log, Plan Archive, File-Session Links, 6 built-in Themes, App Zoom, Configuration Profiles, Prompt Cache, and an 8-step Onboarding Walkthrough.

Can I build plugins for NeuroNest?

Yes. The NeuroNest Plugin SDK (planned) will let developers build custom tools, skills, and integrations using a TypeScript-first SDK. Plugins run locally within the same security sandbox as built-in agents and declare their required capabilities in a manifest.

How do I get access to the beta?

Download NeuroNest at neuronest.cc/download and enter a PQC invite code (format: PQC-XXXXXXXX-XXXXXXXX). Invite codes are derived using post-quantum cryptography (ML-KEM-768, ML-DSA-65, or SLH-DSA-SHAKE-128s). Request a code from an existing user or the NeuroNest team.

Is NeuroNest raising investment?

Yes. NeuroNest has a Series A open. Visit neuronest.cc/investors for the full investor deck covering the problem, solution, market opportunity ($50B+ AI developer tools market), traction, roadmap, team, and the ask.

NeuroNest Security Architecture — 7-Layer Defense-in-Depth

NeuroNest protects every AI agent action with seven independent security layers. No single point of failure. Defense in depth means each layer catches threats that the others miss.

The 7 Security Layers

Firewall Engine — Zero-dependency TypeScript engine running before any text reaches an agent. Applies sanitization, regex-based prompt injection detection, secrets scanning, and policy enforcement in under 1 ms.
Enhanced Firewall — LLM-powered semantic analysis layer catches sophisticated injection attempts that evade pattern matching. Includes configurable policy presets and PII redaction.
Action Analyzer — Pre-execution screening for every shell command, file write, and network request. Pattern Analyzer and Policy Rail Analyzer combine into an ensemble risk score. High-risk actions are blocked; uncertain actions escalate to human approval.
Runtime Protection — Anti-tamper monitoring: DevTools attachment detection, debugger flag scanning, SHA-256 file integrity verification, environment anomaly detection, and keyboard shortcut blocking.
HMAC Request Signing — All inter-agent communication signed with HMAC-SHA256. Nonce-based replay protection prevents message reuse attacks.
Native C++ Crypto Layer — Post-quantum cryptographic operations (ML-KEM-768, ML-DSA-65, SLH-DSA-SHAKE-128s) run in native code outside the LLM execution sandbox.
Edit Lock System — File and directory-level protection. Locked paths cannot be modified by any agent, including the orchestrator, without explicit human approval.

Local-First by Design

Source code is processed locally by default. Nothing is sent to NeuroNest servers. When cloud providers are configured, code goes only to that provider under their privacy policy. You can audit every agent action through the Action Log.

Prompt Injection Protection

NeuroNest's dual-layer firewall is the most comprehensive prompt injection defense available in any IDE. Layer 1 catches known injection patterns in under 1 ms. Layer 2 uses LLM semantic analysis to catch novel injections that pattern matching cannot detect.