FAQ
Answers about Neurova's features and day-to-day usage, so you can get moving quickly.
Neurova is a multi-agent AI framework built on the CogArch 2.0 cognitive architecture. Unlike conventional conversational AI, Neurova gives every agent its own "brain" (memory and personality) while sharing a "cerebellum" (plan orchestration) and a "brainstem" (execution engine) to handle tasks efficiently and keep evolving its capabilities.
Neurova is not simply a chatbot but an autonomous agent framework with a complete closed-loop cognitive architecture. Its hallmarks are:
Independent memory system: five layers — working, short-term, long-term, emotional memory and a knowledge graph
Personality and growth: personality evolution, motivation-driven behaviour, reflective learning and meta-cognitive optimization
Multi-agent collaboration: sequential, parallel, master-slave and consensus modes
Security: a built-in three-layer agent firewall that protects your data
Independent memory system: five layers — working, short-term, long-term, emotional memory and a knowledge graph
Personality and growth: personality evolution, motivation-driven behaviour, reflective learning and meta-cognitive optimization
Multi-agent collaboration: sequential, parallel, master-slave and consensus modes
Security: a built-in three-layer agent firewall that protects your data
CogArch 2.0 is Neurova's core cognitive architecture, modelled on the functional regions of a human brain:
Cerebral cortex: observing information, recalling memory, logical reasoning, deciding actions, self-reflection
Cerebellum: decomposing intent, generating tasks, orchestrating execution, evaluating results
Brainstem: calling tools, running workflows, scheduling resources, monitoring execution
Spinal cord: dispatching events, connecting modules, attaching external channels
The full cognitive cycle: observe → recall → reason → act → reflect and consolidate.
Cerebral cortex: observing information, recalling memory, logical reasoning, deciding actions, self-reflection
Cerebellum: decomposing intent, generating tasks, orchestrating execution, evaluating results
Brainstem: calling tools, running workflows, scheduling resources, monitoring execution
Spinal cord: dispatching events, connecting modules, attaching external channels
The full cognitive cycle: observe → recall → reason → act → reflect and consolidate.
Per the installation guide:
Requirements: Python 3.10+ (the project's own venv is recommended), Node.js 18+ for the frontend, SQLite 3 (FTS5 full-text search included), and at least 4GB of memory
Install steps: clone the repository → pip install -r requirements.txt → copy .env.example to .env and fill in your LLM API key → npm install in the NeurUI directory
Start: python start.py --backend (backend on 9527) or --chat (everything), then verify at http://localhost:9527/health
Requirements: Python 3.10+ (the project's own venv is recommended), Node.js 18+ for the frontend, SQLite 3 (FTS5 full-text search included), and at least 4GB of memory
Install steps: clone the repository → pip install -r requirements.txt → copy .env.example to .env and fill in your LLM API key → npm install in the NeurUI directory
Start: python start.py --backend (backend on 9527) or --chat (everything), then verify at http://localhost:9527/health
Yes. Per the Docker deployment guide:
Quick start: docker pull neurova/neurova:latest and then docker run, mounting the host's data/, config/ and logs/ directories into the container and injecting secrets through the NEUROVA_DB_PASSWORD and NEUROVA_SESSION_SECRET environment variables
docker-compose: an official compose file is provided (ports 80/443, restart: unless-stopped), suited to one-command production deployment
Quick start: docker pull neurova/neurova:latest and then docker run, mounting the host's data/, config/ and logs/ directories into the container and injecting secrets through the NEUROVA_DB_PASSWORD and NEUROVA_SESSION_SECRET environment variables
docker-compose: an official compose file is provided (ports 80/443, restart: unless-stopped), suited to one-command production deployment
Neurova ships a complete protection stack:
L0 - entry gateway: JWT authentication, IP allow/deny lists, rate limiting, input safety inspection and automatic masking on output
L1 - user and agent isolation: data is fully isolated between users, and agents under the same user can be isolated as configured
L2 - file access protection: sensitive-file interception, workspace allowlists and custom path rules
L0 - entry gateway: JWT authentication, IP allow/deny lists, rate limiting, input safety inspection and automatic masking on output
L1 - user and agent isolation: data is fully isolated between users, and agents under the same user can be isolated as configured
L2 - file access protection: sensitive-file interception, workspace allowlists and custom path rules
Neurova implements a layered, human-like memory system:
Working memory: the current conversation context, updated in real time
Short-term memory: recent events and conversations, with a temperature decay mechanism
Long-term memory: experience and knowledge persisted under version control
Emotional memory: memories tagged with affect, which steer reasoning and decisions
Knowledge graph: a graph of events and entities supporting causal reasoning
Distinctive capabilities include temperature decay, high-temperature memory extraction, memory consolidation and meta-cognitive monitoring.
Working memory: the current conversation context, updated in real time
Short-term memory: recent events and conversations, with a temperature decay mechanism
Long-term memory: experience and knowledge persisted under version control
Emotional memory: memories tagged with affect, which steer reasoning and decisions
Knowledge graph: a graph of events and entities supporting causal reasoning
Distinctive capabilities include temperature decay, high-temperature memory extraction, memory consolidation and meta-cognitive monitoring.
Yes. Neurova has built-in LLM provider management and automatic LLM routing:
Provider management: add multiple providers (OpenAI and others), set default models, and configure everything from either the admin console or the CLI
Automatic routing: the LLM Router matches model capability to the request type, covering ten request types including text chat, image/audio/video understanding, text-to-image, image-to-image, text-to-video, image-to-video, and speech synthesis and recognition
Selection algorithm: capability filtering → health check → load and priority, so each request lands on the best-suited model
Provider management: add multiple providers (OpenAI and others), set default models, and configure everything from either the admin console or the CLI
Automatic routing: the LLM Router matches model capability to the request type, covering ten request types including text chat, image/audio/video understanding, text-to-image, image-to-image, text-to-video, image-to-video, and speech synthesis and recognition
Selection algorithm: capability filtering → health check → load and priority, so each request lands on the best-suited model
Yes. The knowledge base module offers an isolation-and-sharing model across multiple users and multiple knowledge bases, with hybrid RAG retrieval:
Four visibility tiers: public, private (creator only, the default), shared_with (specific users or groups), and approval-gated (for sensitive knowledge bases)
Retrieval pipeline: the user's question goes through hybrid retrieval (vectors + FTS5 full text + RRF fusion ranking), then permission filtering, then context fusion to generate the answer
Four visibility tiers: public, private (creator only, the default), shared_with (specific users or groups), and approval-gated (for sensitive knowledge bases)
Retrieval pipeline: the user's question goes through hybrid retrieval (vectors + FTS5 full text + RRF fusion ranking), then permission filtering, then context fusion to generate the answer
Neurova has a complete skill ecosystem:
Built-in skills: baseline capabilities shipped with the framework (text, code, data, knowledge and more)
Public skill pool: a marketplace model with search, download and install
Evolution engine: skills improve automatically with usage frequency and feedback
Experience knowledge base: every successful run has its experience extracted and stored automatically
Built-in skills: baseline capabilities shipped with the framework (text, code, data, knowledge and more)
Public skill pool: a marketplace model with search, download and install
Evolution engine: skills improve automatically with usage frequency and feedback
Experience knowledge base: every successful run has its experience extracted and stored automatically
Neurova unifies access across many message channels:
REST API: a standard HTTP interface
WebSocket: real-time two-way communication
Telegram: instant messaging
WeChat: official accounts and WeCom
Discord: community servers
Custom webhooks: callbacks from any external system
All channels share one message format, with SSE streaming for real-time output.
REST API: a standard HTTP interface
WebSocket: real-time two-way communication
Telegram: instant messaging
WeChat: official accounts and WeCom
Discord: community servers
Custom webhooks: callbacks from any external system
All channels share one message format, with SSE streaming for real-time output.
The Neurova agent growth system comprises:
Personality: a Big Five model with custom traits and evolution triggers
Motivation: three intrinsic drives — curiosity, achievement and sociability
Reflection: periodic reflection logs, a question queue and proactive questioning
Meta-cognition: self-monitoring, reflection and optimization, including health checks of the memory system
Constitution: behavioural rules and ethical constraints, with custom rules supported
Personality: a Big Five model with custom traits and evolution triggers
Motivation: three intrinsic drives — curiosity, achievement and sociability
Reflection: periodic reflection logs, a question queue and proactive questioning
Meta-cognition: self-monitoring, reflection and optimization, including health checks of the memory system
Constitution: behavioural rules and ethical constraints, with custom rules supported
Neurova simulates a human sleep-wake cycle to manage resources intelligently:
Beta (awake): full functionality, normal responsiveness (13-30 Hz)
Alpha (light sleep): reduced frequency, low-priority tasks (8-13 Hz)
Theta (deep deliberation): memory sorting and consolidation (4-8 Hz)
Gamma (REM): creative association and dream generation (30+ Hz)
Automatic sleep, manual control, brainwave visualization and scheduled wake-up are all supported.
Beta (awake): full functionality, normal responsiveness (13-30 Hz)
Alpha (light sleep): reduced frequency, low-priority tasks (8-13 Hz)
Theta (deep deliberation): memory sorting and consolidation (4-8 Hz)
Gamma (REM): creative association and dream generation (30+ Hz)
Automatic sleep, manual control, brainwave visualization and scheduled wake-up are all supported.
See the troubleshooting guide:
Database connection fails: check permissions on the data/ directory (chmod 755 data/) and confirm the database file exists and is writable
Cannot log in to the admin console: confirm cookies are enabled in the browser, check permissions on the session directory, then clear the cache and retry
API returns 401: verify the Authorization header format and the key's status and validity period
Where the logs are: application log at logs/app.log, access log at logs/access.log; more remedies are in the "Troubleshooting" article in the documentation center
Database connection fails: check permissions on the data/ directory (chmod 755 data/) and confirm the database file exists and is writable
Cannot log in to the admin console: confirm cookies are enabled in the browser, check permissions on the session directory, then clear the cache and retry
API returns 401: verify the Authorization header format and the key's status and validity period
Where the logs are: application log at logs/app.log, access log at logs/access.log; more remedies are in the "Troubleshooting" article in the documentation center
Still have questions?
Reach out through any of the channels below