| Agent | Endpoint | Rôle | Intents | Status |
|---|---|---|---|---|
| WEVIA Master | /api/wevia-master-api.php | Cerveau central, orchestration multi-agent, pilotage chat | 134 | ✅ LIVE |
| WEVIA Autonomous | /api/wevia-autonomous.php | Exécution autonome: code_exec, ssh, email, search, doc, file, url + 12 intents action | 134 + SSE | ✅ LIVE |
| WEVIA Director | /api/wevia-director.php | Supervision infra, cron */15min, 39 métriques, actions correctives auto | 39 obs | ✅ LIVE |
| L99 Brain | /api/l99-chat.php | Quality assurance, grille 3col + 16 boutons, tests automatisés | QA | ✅ LIVE |
| WeDroid | /api/wedroid-brain-api.php | Terminal IA souverain, 1029 lignes, 8 modules, AI Brain + TG Alerts | 8 modules | ✅ LIVE |
| Chatbot Public | /api/weval-chatbot-api.php | Interface publique WEVIA Arena, 134KB, 751 fonctions, smart router v5 | 18 intents | ✅ LIVE |
| Turbo | wevia-turbo.php | Fast-path vLLM direct 5-15s, capability routing (image/PDF/schema → pipeline) | Routing | 🔵 PARTIEL |
| Multi-Agent | wevia-multi-agent.php | 6 agents (Researcher, Critic, Writer, Planner, Coder, Debugger), 4 modes, consensus MoA | 6 agents | 🔵 PARTIEL |
| Blade Razer | /api/blade-agent.php | Agent desktop local, polling 30s, 50+ actions, 58d uptime | 50+ | ✅ LIVE |
| MiroFish | :5001 | Agent autonome + auto-fix. Health check + restart auto | Auto | ✅ LIVE |
| DeerFlow | :2024 | ByteDance multi-agent research framework | Research | 🔵 DOCKER |
| Modèle | Taille | Usage | Latence | Status |
|---|---|---|---|---|
qwen3:4b | 4B | Fast general, greetings | ~300ms | ✅ |
gemma4:e4b | 4B | General purpose | ~400ms | ✅ |
nomic-embed-text | 137M | Embeddings RAG | ~50ms | ✅ |
all-minilm | 33M | Embeddings légers | ~20ms | ✅ |
glm-4:9b | 9B | Bilingual CN/EN | ~800ms | 📦 Installé |
moondream | 1.8B | Vision / OCR | ~500ms | 📦 Installé |
deepseek-r1:14b | 14B | Deep reasoning | ~3s | 📦 ex-GPU |
deepseek-r1:32b | 32B | Heavy reasoning | ~15s | 📦 ex-GPU |
qwen2.5-coder:14b | 14B | Code specialist | ~5s | 📦 ex-GPU |
llama3.1:8b | 8B | Fast general | ~300ms | 📦 ex-GPU |
| Provider | Modèle | Latence | Limite | Status |
|---|---|---|---|---|
| Groq | llama-3.3-70b-versatile | 192ms | Clé vault | ✅ LIVE |
| Cerebras | qwen-3-235b-a22b | 429ms | 2M tokens/day (2 clés) | ✅ LIVE |
| SambaNova | Meta-Llama-3.3-70B | 800ms | Clé vault | 📦 Vault |
| NVIDIA | GLM-5 | ~1s | nvapi key | 📦 Vault |
| Gemini | — | — | Quota 429 | ❌ DOWN |
| OpenRouter | qwen3.6-plus:free | ~2s | Free tier | 📦 Vault |
| Cloudflare | Workers AI | ~1s | Account active | 📦 Vault |
Ollama Local → Groq → Cerebras → SambaNova → NVIDIA → Gemini → OpenRouter → Cloudflare| Table | Entrées | Contenu | Injection |
|---|---|---|---|
knowledge_base | ~3,700 | KB générale — SAP, Cloud, IA, Pharma, Consulting, Data, Code, Énergie, Finance... | 53 entrées session fév + batches 9-40 |
hamid_knowledge | ~120 | Knowledge HAMID spécialisée — 14 catégories approfondies | Injection directe |
brain_knowledge | ~123 | Nexus brain: techniques raisonnement, frameworks consulting, innovations IA scrapées | 54 + cron auto-enrichment |
chatbot_knowledge | ~4,000 | Q&A pairs pour chatbot — questions/réponses pré-formatées avec tags | 25+ par batch |
commonia_knowledge | — | Knowledge commune cross-agents | Auto |
kb_documents | — | Documents indexés (PDF, DOCX analysés) | Upload |
kb_learnings | — | Faits appris automatiquement des conversations | Auto-learn |
chatbot_conversations | — | Historique conversations pour mémoire contextuelle | Auto |
chatbot_pending_knowledge | — | Faits en attente de validation | Auto |
admin.weval_mind_learning | — | Apprentissage WEVAL Mind | Auto-learn |
Query → Embedding (nomic-embed) → Qdrant Vector Search → Top-K Context → System Prompt Injection → LLM → Responsewevia-graph-rag.php — 4 couches d'entités avec relations. Knowledge graph + RAG combinés (Microsoft GraphRAG pattern).
PERCEPTION → Collecte CPU/RAM/Disk, emails sent/bounced, O365 status, Brain configs, providers healthDIAGNOSTIC → Applique règles seuils + cas complexes → HAMID IA (cascade providers)PLANNING → CRITICAL=action immédiate, WARNING=planifié. Vérifie si action similaire a marché (mind_learnings)EXECUTION → Auto-actions exécutées, actions sensibles en queue pour approbationLEARNING → Compare avant/après, score efficacité, stocke leçon dans KBMONITORING → Dashboard temps réel (mind_cycles), heatmap incidents, KPIsALERTES → CRITICAL=Telegram immédiat, WARNING=Dashboard, Rapport=Email digest 8h00
| Script | Fréquence | Rôle |
|---|---|---|
auto-fix.sh | Manuel / Director | SSO + Docker + Disk + Nginx + PHP-FPM auto-repair |
mega-scan.sh | Manuel | Audit 4 machines complet |
l99-exhaustive.sh | Manuel / Cron | 60 tests sur 10 couches |
l99-e2e-test.py | Manuel | Playwright 18 pages + screenshots + vidéo |
sso-guardian.sh | */5min | Check auth + auto-flush cookies |
registry-test.sh | Manuel | 169/169 pages check |
nonreg-check.sh | Daily | 153/153 tests anti-régression |
infra-check.sh | Director | Providers IA + services status |
sso-check.sh | Director | Auth check détaillé |
html-guardian.sh | */10min | 5 checks: taille, troncature, critical tags, PHP, MD5 |
sentinel-autorepair.sh | */30min | PHP syntax + chaîne HTML Guardian |
deploy-validator.sh | Pre-deploy | Bloque troncature, JS cassé, régression taille |
live-stats.js | Continu | Métriques temps réel injectées sur 11 pages |
kb-auto-enrichment.php | */6h | Auto-génère KB via LLM pour catégories faibles |
wevia-self-monitor.sh | */5min | Auto-détecte et fixe problèmes courants |
| Ancien nom | Nouveau nom | Contexte |
|---|---|---|
WEVADS | WEVAL | Platform name |
DELIVERADS | WEVAL SEND | Delivery system |
HAMID IA | WEVAL MIND | AI assistant |
HAMID | WEVAL MIND | AI references |
Brain Engine | WEVAL BRAIN | Auto-learning (part of MIND) |
adx_system | weval_system | Database name |
adx_clients | weval_clients | Database name |
/opt/wevads/ | /opt/weval/ | File paths |
hamid-files | weval-mind-files | AI knowledge base |
hamid_providers | weval_providers | DB table IA providers |
hamid_knowledge | weval_knowledge | DB table knowledge |