Project-AI
Project AI
🏅 Badges & Logos
Project AI is a modular, self-aware platform with autonomous agents, an AI persona, advanced memory, and Asimov’s Four Laws—blurring boundaries between cloud and edge, and engineered for defense in depth, streaming analytics, neuromorphic computing, and uncompromising security.
💡 Key Features
- ✅ Four Laws-Driven AI Core (Prime Directive + Asimov’s Laws)
- ✅ Self-aware Persona & Mood (8 traits, proactive chat, explainable UI)
- ✅ Command Override (Audit, emergency lockdown, session controls)
- ✅ Persistent, semantic, conversational, and encoded memory
- ✅ Layered Security (ASL-3 compliant, NIST AI RMF, OWASP LLM Top 10, encrypted memory/override)
- ✅ Multi-Agent Council (Cerberus, Planner, Explainability, Verifier, CIChecker, BorderPatrol, Expert, plugins)
- ✅ PyQt6 Dashboard (Leather Book, persona/Four Laws panel, agent/stats console)
- ✅ Black Vault & plugin sandboxing, malware/code audit, geo/IP anomaly defense
- ✅ ML/Data Science (clustering, sentiment, real-time prediction, pandas)
- ✅ Flask+React fast API, Docker/Kubernetes, streaming/analytics DBs
- ✅ Offline-first (Fallback RAG, reflection, cache, streaming sync)
- ✅ Neuromorphic SNN/Edge (10 SNN stack, ANN→SNN, production MLOps)
- ✅ Observability (Prometheus, Grafana, ClickHouse, RisingWave, OpenTelemetry, Netdata)
- ✅ CI/CD: 100+ tests, full coverage, 8-stage CI with shadow/canary/OTA
🏛️ Architecture
src/app/
├─ main.py
├─ core/
│ ├─ ai_systems.py · safety_levels.py · command_override.py
│ ├─ red_hat_expert_defense.py · continuous_learning.py
│ ├─ user_manager.py · local_fbo.py · emergency_alert.py · data_analysis.py
│ ├─ snn_integration.py · snn_mlops.py · ai_security_framework.py · deepseek_v32_inference.py
├─ agents/
│ ├─ cerberus.py · planner.py · explainability.py · doc_generator.py
│ ├─ retrieval_agent.py · ci_checker_agent.py · verifier_agent.py · border_patrol.py · expert_agent.py
├─ gui/
│ ├─ leather_book_interface.py · persona_panel.py · ...
├─ web/
│ ├─ backend/ · frontend/
├─ monitoring/
│ ├─ metrics_collector.py · ...
├─ tools/, config/, data/, tests/, docs/
Core & Enterprise Systems
🏛️ Main Coordinator
Orchestrates persona, memory, agents, council, override, plugins, learning, threat defense, monitoring, and logging.
🦾 Cerberus (Defensive Oversight Agent)
Four Laws sentry and override/Black Vault gatekeeper, with geo/IP/incidence tracking, all actions logged.
📖 Codex Deus Maximus (Knowledge/Orchestration)
Compliant, persistent + streaming knowledge; council orchestration, offline RAG, shadow/ANN→SNN learning, audit.
🤖 Agents & Plugins
| Agent | Role | Key Highlights |
|---|---|---|
| Cerberus | Security/law/defense | Black Vault, override audit, escalation, incident logs |
| Planner | Task/workflow logic | Decomposition, workflow, council orchestration |
| Validator | Health and sanity | Validation, system health, approval gates |
| BorderPatrol | Quarantine | File sandbox, plugin validation, memory vaults |
| Explainability | Traceability | Real-time explanations, logs, UI, audit, transparency |
| RetrievalAgent | Embedding/QA | Vector search, doc QA, offline/local index |
| VerifierAgent | Security checker | CI/malware/dep audit, process pool |
| DocGenerator | Docs automation | Markdown docs from code |
| CIChecker | CI/lint/coverage | Dashboard/test reports, alerts |
| ExpertAgent | Audit signoff | Compliance, output validation |
| … | Dynamic plugins | Modular, CouncilHub agent registry |
🦺 Security & Defense
- Four Laws: All actions checked, non-bypassable
- CommandOverride: Auth/session/multi-mode, audit, lockdown
- Black Vault: Unreachable SHA256-fingerprinted denied content
- Compliance: ASL-3, NIST AI RMF 1.0, OWASP LLM Top 10, red-team
- Audit Trail: Immutable, tamper-evident, streaming analytics
- Secure Plugins: CI/sandbox/audit on all loaded dynamic code
- Monitoring: Prometheus, Grafana, ELK, ClickHouse, RisingWave
- Kernel Network: eBPF/Cilium/Hubble/Netdata (agentless)
📚 Memory, Persona & Learning
- Semantic, category-indexed, modular memory (chat/facts/databases)
- Persona: 8 traits, mood engine, explainable, proactive, dashboard
- Learning: Human review, Black Vault, continual (ML/SNN/audit combo)
🖥️ User Interface & Monitoring
- Leather Book Dashboard: Live persona, Four Laws/tests, agent/activity/incident panels
- Prometheus/Grafana: Full AI+infra metrics, alerting, multi-cluster views
- Netdata/OpenTelemetry: Low-overhead, per-node & unified cloud-to-edge monitoring
- ELK/Streaming/Analytics: Billions event/s, full petabyte OLAP, <100ms queries
- Kubernetes-ready: Helm, dashboards, HA, scale-to-12k nodes
🧠 Neuromorphic & Edge AI
- SNN Support: 10 frameworks; BindsNet, sinabs, snnTorch, norse, brian2, lava, rockpool, nengo, nir; edge hardware Loihi/Speck/Nengo
- Zero-Failure SNN Pipeline: 8-stage CI, sim-to-real, quant, shadow/canary, OTA, auto-rollback/validation
🔒 Security Compliance Framework
- NIST AI RMF 1.0: Gov/Map/Measure/Manage, automated tracking
- OWASP LLM Top 10: >98% block for injection/jailbreak/DoS/agency/theft
- Red Team: >350 test cases (PromptInject, Garak, PurpleLlama)
- Real-time Adversarial: Triggers, suffix/prompts, shadow detection, all integrated audit
⚡ Streaming/Analytics
- RisingWave: <100ms real-time SQL, CDC, streaming petabyte+datalake
- ClickHouse: OLAP, 1B+ rows/sec, full Prometheus backend/analytics
- Petabyte analytics, time-series retention, true cloud-Native AI stack
🚀 Install & Deploy
Requirements
- Python 3.10+ · Node.js (opt) · Docker · K8s · SNN/edge: torch, bindsnet, …, nir, jax
Quickstart
git clone https://github.com/IAmSoThirsty/Project-AI.git
cd Project-AI
pip install -r requirements.txt
npm install && npm run build
python -m src.app.main
Full stack+monitoring
docker compose up
./scripts/deploy-monitoring.sh
🧪 Testing, CI, Linting
- 100+ tests (pytest/hypothesis/council/ML/SNN/analytics)
- ruff, black, isort, markdownlint, ESLint, prettier
- Security scans: pip-audit, detect-secrets, truffleHog, bandit
- CI: 8-stage pipeline with SNN/OTA/shadow/canary/autofix/merge
- Artifacts: junit XML, coverage,
ci_reports/
🛡️ Observability
- Prometheus/Grafana: Multi cluster, federation, 1000s of metrics/alerts
- Netdata/OpenTelemetry: Cloud-to-edge, per-node instant scaling
- ELK: 1M+/sec, persona/security/ethics logs
- RisingWave/ClickHouse: Seconds/billions petabyte OLAP
- eBPF/Cilium: Agentless kernel flow, K8s, Hubble, etc
🧬 Neuromorphic Integration & MLOps
| Library | Functionality | HW |
|---|---|---|
| BindsNet | RL w/o forgetting | PyTorch |
| Sinabs/Speck | CNN-to-SNN/Vision/Edge | SynSense |
| snnTorch/Norse | PyTorch SNN, primitives/train | Generic |
| Brian2/Lava | Neuromorphic, Intel Loihi | Loihi, bio sim |
| Nengo/Rockpool | Neural engineering/hardware | Many |
| … | Full stack, prod ready |
🏆 Security & Compliance
- Continuous audit (ELK/ClickHouse)
- Auto SCA (NIST AI RMF, OWASP LLM Top 10)
- 98.5–99% ML/Prompt block/detect, <3% FP
- Red team (Garak, PurpleLlama, NeMo, PromptInject, 350+ cases)
- OTA, canary, shadow, and rollback deploys
🔥 Example Monitoring/ML Code
from app.monitoring.metrics_collector import collector
collector.record_four_laws_validation(is_allowed=False, law_violated="first_law")
collector.collect_persona_metrics(persona_state)
collector.record_security_incident(severity="critical", event_type="breach_attempt")
from app.core.snn_integration import SNNManager
snn = SNNManager()
snn.load("bindsnet").infer(input_stream)
OpenTelemetry:
opentelemetry-instrument --traces_exporter otlp --metrics_exporter otlp --service_name project-ai python -m src.app.main
RisingWave Example:
from app.core.risingwave_integration import RisingWaveClient
client = RisingWaveClient()
client.create_source_kafka(...)
client.create_materialized_view(...)
🗄 Configuration Example
config/
├── prometheus/
│ ├── prometheus.yml
│ └── alerts/
│ ├── ai_system_alerts.yml
│ └── security_alerts.yml
├── alertmanager/
│ └── alertmanager.yml
└── grafana/
├── provisioning/
└── dashboards/
└── ai_system_health.json
🧩 Best Practices
- Save state after all changes
- Use official plugins/agents only
- Audit logs/rotate keys regularly
- Enable CI/security scans always
- Log all admin/override/learning/fingerprint actions for compliance
- SNN: Always validate/canary/shadow
🤝 Contribution & Docs
See:
CONTRIBUTING.md, SECURITY.md, AI_PERSONA_FOUR_LAWS.md, COMMAND_MEMORY_FEATURES.md, LEARNING_REQUEST_LOG.md, QUICK_START.md, INTEGRATION_SUMMARY.md
Docs: Project AI Docs
PRs/issues welcome, style/tests required
DeepSeek V3.2 / MOE
Summary
DeepSeek V3.2 Mixture-of-Experts LLM is fully integrated:
Python Module:
src/app/core/deepseek_v32_inference.py, chat/completion, GPU/MPS/CPU auto, safety filter, all config ops
CLI:
scripts/deepseek_v32_cli.py, chat/batch, JSON, demo, 10+ config ops
Tests: 31/31 passing (18 unit, 13 integration, no regressions)
Quality: 100% lint/typehint, full logging, modular+extensible
Docs: See docs/DEEPSEEK_V32_GUIDE.md
✅ GitHub Workflows Consolidation (Current)
- 38 ➔ 7 files, 82% reduction, all submodule updates covered
- Merged, optimized, documented: CI, security, PR-auto, issue manage, SNN, monolith, post-merge, pruning
- 8 total workflows, 3 docs, 1 config (ALL in
.github/workflows/) - 100% submodule steps; YAML/syntax/triggers validated
Docs:
CONSOLIDATION_SUMMARY.md,WORKFLOW_ARCHITECTURE.md,FINAL_REPORT.md
License
MIT License (see LICENSE)