🌐 Buzzy Navigator
Service agent-browser BPA-native — fetch URL → annotation 6 primitives + scoring 23 facets + plan agent GLM-5.1 LIVE .
LIVE production
zai/glm-5.1 ✓
v0.1.0-pre.2
cross-LLM (zai|anthropic|openai|ollama)
@bfw/browser-bpa-layer
@bfw/navigator
Tauri 2 desktop alt
23/23 facets coverage
194 tests pass
📡 Test live — analyse + plan agent
URL
Intent
🚀 Analyser + planifier (GLM-5.1)
🛠 API endpoints
GET /health — status liveness probe
GET / — cette page (landing interactive)
POST /analyze body {"url":string,"intent"?:string} → JSON complet (annotation + score + plan)
🧪 Test depuis terminal
curl -X POST https://navigator.buzzyai.fr/analyze \
-H "Content-Type: application/json" \
-d '{"url":"https://example.com","intent":"explorer"}'
🧬 Pipeline interne
1. fetch URL HTML brut (~500-1000 ms)
2. annotateDom → 6 primitives BPA (Node + Capability + Contract + Intent + Context + Signal)
3. scoreAnnotation → 23 facets scoring agent-readability
4. describeAnnotation L0 (~5 tokens) + L1 (~50 tokens) — progressive disclosure
5. LLM GLM-5.1 plan agent JSON structured (5-15s) — Z.AI Codeplan
🎯 Vision BPA appliquée
6 primitives Décomposition page web en briques sémantiques agent-friendly
23 facets Scoring multi-dimensionnel agent-readability (shape/auth/perf/privacy/error/...)
L0→L3 Progressive Disclosure token-efficient (5 → 50 → 500 → runtime full)
BpaResult Failure-as-feature ADR-031 — jamais throw, toujours plan B + alternative
Visible AI Chaque étape transparente (annotation + score + plan + LLM raw)