feat: real router + live 98-model registry

app.py

@@ -1,301 +1,173 @@
-"""MaramaRoute Live — interactive sovereign router demo (v2 polish)."""
-from __future__ import annotations
-
+"""
+MaramaRoute live demo — runs the real router against the live 98-model
+registry pulled from AbteeXAILab/marama-route on Hugging Face.
+"""
 import json
+import os
 from dataclasses import dataclass, field
-from pathlib import Path
-from typing import Any, Dict, List, Optional
+from typing import Any, Optional
 
 import gradio as gr
 from huggingface_hub import hf_hub_download
 
+REGISTRY_REPO = "AbteeXAILab/marama-route"
+REGISTRY_PATH = "configs/lumynax_model_registry.json"
 
-# ---------- Embedded router ----------
-@dataclass
-class RouteDecision:
-    selected: Optional[Dict[str, Any]]
-    fallbacks: List[Dict[str, Any]] = field(default_factory=list)
-    rejected: List[Dict[str, Any]] = field(default_factory=list)
-    reasons: List[str] = field(default_factory=list)
-
-    def to_dict(self) -> Dict[str, Any]:
-        return {"selected": self.selected, "fallbacks": self.fallbacks, "rejected": self.rejected, "reasons": self.reasons}
-
-
-_TASK_TO_TAGS = {
-    "code": {"coder", "code"}, "reasoning": {"reasoning"},
-    "multimodal": {"multimodal", "image", "audio", "voice", "vision"},
-    "embedding": {"embedding", "embed", "retrieval"}, "chat": {"general", "instruct"},
-}
-
-
-def _filter(model: Dict[str, Any], request: Dict[str, Any]) -> Optional[str]:
-    req_mod = set(request.get("modalities", []))
-    mod = set(model.get("modalities", []))
-    if req_mod and not req_mod.issubset(mod):
-        return f"modalities missing: needs {sorted(req_mod)}, has {sorted(mod)}"
-    if request.get("min_context_tokens") and (model.get("context_tokens") or 0) < request["min_context_tokens"]:
-        return f"context_tokens {model.get('context_tokens')} < required {request['min_context_tokens']}"
-    if request.get("jurisdiction") and request["jurisdiction"] not in (model.get("residency") or []):
-        return f"residency {model.get('residency')} does not include {request['jurisdiction']}"
-    if request.get("requires_local") and model.get("runtime") not in ("llama_cpp", "llama_cpp_multimodal", "transformers", "transformers_multimodal", "python_embedding"):
-        return f"requires_local but runtime is {model.get('runtime')}"
-    if request.get("requires_tools") and not model.get("supports_tools", False):
-        return "supports_tools = false"
-    if request.get("requires_json") and not model.get("supports_json", False):
-        return "supports_json = false"
-    if str(request.get("data_sensitivity", "")).lower() in ("restricted", "personal", "health", "iwi", "taonga"):
-        if (model.get("sovereignty_tier") or 0) < 2:
-            return f"sovereignty_tier {model.get('sovereignty_tier')} < 2 for sensitive data"
-    return None
-
-
-def _score(model: Dict[str, Any], request: Dict[str, Any]) -> float:
-    q = 6 - int(model.get("quality_rank") or 6)
-    c = 6 - int(model.get("cost_rank") or 6)
-    s = int(model.get("sovereignty_tier") or 0)
-    active = float(model.get("active_params_b") or model.get("total_params_b") or 0)
-    base = 2.0 * q + 1.5 * s + 0.5 * c
-    wanted = _TASK_TO_TAGS.get(str(request.get("task_type", "")).lower(), set())
-    if wanted & set(model.get("tags") or []):
-        base += 3.0
-    if request.get("requires_local") and model.get("runtime") == "llama_cpp":
-        base += 1.0
-    if active and active > 50:
-        base -= 0.5
-    return round(base, 3)
-
-
-def route(registry: Dict[str, Any], request: Dict[str, Any]) -> RouteDecision:
-    candidates, rejected = [], []
-    for m in registry.get("models", []):
-        why = _filter(m, request)
-        if why:
-            rejected.append({"repo_id": m.get("repo_id"), "reason": why})
-        else:
-            candidates.append({**m, "_score": _score(m, request)})
-    candidates.sort(key=lambda x: x["_score"], reverse=True)
-    if not candidates:
-        return RouteDecision(None, [], rejected, ["no candidate satisfies the gates"])
-    sel = candidates[0]
-    max_fb = int(request.get("max_fallbacks", 3))
-    return RouteDecision(sel, candidates[1 : 1 + max_fb], rejected, [f"selected {sel['repo_id']} (score {sel['_score']})"])
-
-
-REGISTRY_PATH = Path(__file__).parent / "lumynax_model_registry.json"
-
-
-def load_registry() -> Dict[str, Any]:
-    if REGISTRY_PATH.exists():
-        return json.loads(REGISTRY_PATH.read_text(encoding="utf-8"))
-    path = hf_hub_download(repo_id="AbteeXAILab/marama-route", filename="configs/lumynax_model_registry.json", repo_type="model")
-    return json.loads(Path(path).read_text(encoding="utf-8"))
-
-
-REGISTRY = load_registry()
-MODEL_COUNT = len(REGISTRY.get("models", []))
-
-
-EXAMPLE_CODE = {
-    "prompt": "Refactor this private Python service and explain the diff.",
-    "task_type": "code", "modalities": ["text"], "jurisdiction": "NZ",
-    "data_sensitivity": "restricted", "min_context_tokens": 4096,
-    "requires_local": True, "requires_tools": False, "requires_json": True, "max_fallbacks": 3,
-}
-EXAMPLE_REASON = {
-    "prompt": "Plan a multi-step migration of a legacy data pipeline.",
-    "task_type": "reasoning", "modalities": ["text"], "jurisdiction": "NZ",
-    "data_sensitivity": "restricted", "min_context_tokens": 8192,
-    "requires_local": True, "requires_tools": True, "requires_json": True, "max_fallbacks": 3,
-}
-EXAMPLE_MULTIMODAL = {
-    "prompt": "Describe this image and draft a short public caption.",
-    "task_type": "multimodal", "modalities": ["text", "image"], "jurisdiction": "NZ",
-    "data_sensitivity": "public", "min_context_tokens": 4096,
-    "requires_local": False, "requires_tools": False, "requires_json": False, "max_fallbacks": 3,
-}
-EXAMPLE_EMBEDDING = {
-    "prompt": "Index our internal policy corpus.",
-    "task_type": "embedding", "modalities": ["text"], "jurisdiction": "NZ",
-    "data_sensitivity": "restricted", "min_context_tokens": 4096,
-    "requires_local": True, "max_fallbacks": 3,
-}
-EXAMPLE_TINY = {
-    "prompt": "Smoke test on a low-spec laptop.",
-    "task_type": "chat", "modalities": ["text"], "jurisdiction": "NZ",
-    "data_sensitivity": "public", "min_context_tokens": 2048,
-    "requires_local": True, "requires_json": False, "max_fallbacks": 3,
+BRAND_CSS = """
+:root {
+  --lx-paper:#fffefa; --lx-ink:#0a0a0b; --lx-amber:#e08a2c;
+  --lx-amber-dark:#9a5416; --lx-muted:#726b62; --lx-soft:#f6f0e8;
 }
+body, .gradio-container { background: var(--lx-paper) !important; color: var(--lx-ink) !important; }
+h1, h2, h3 { font-family: 'Cormorant Garamond','EB Garamond',Georgia,serif; }
+.lx-hero { background: linear-gradient(135deg,#fffefa 0%,#f6f0e8 100%); border: 1px solid rgba(10,10,11,0.08); border-radius: 12px; padding: 18px 22px; margin-bottom: 14px; }
+.lx-pick { background: #f0f7ee; border-left: 4px solid #4caf50; padding: 12px 14px; border-radius: 6px; }
+.lx-runner-up { background: var(--lx-soft); border-left: 3px solid var(--lx-amber); padding: 10px 12px; border-radius: 6px; margin-top: 8px; }
+table.lx-runners { border-collapse: collapse; width: 100%; font-size: 0.9em; }
+table.lx-runners th, table.lx-runners td { border-bottom: 1px solid rgba(10,10,11,0.08); padding: 6px 8px; text-align: left; }
+"""
 
-
-HERO_SVG = """<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 1280 220" role="img" aria-label="MaramaRoute architecture">
-  <defs><marker id="ar2" viewBox="0 0 10 10" refX="9" refY="5" markerWidth="8" markerHeight="8" orient="auto-start-reverse"><path d="M0 0 L10 5 L0 10 z" fill="#e08a2c"/></marker></defs>
-  <rect width="1280" height="220" fill="#fffefa"/>
-  <rect x="0" y="0" width="1280" height="3" fill="#0a0a0b"/>
-  <text x="64" y="32" font-family="ui-monospace,Menlo,Consolas,monospace" font-size="11" font-weight="700" letter-spacing="0.2em" fill="#9a5416">MARAMAROUTE · SIX-GATE SOVEREIGN ROUTER</text>
-  <g font-family="ui-monospace,Menlo,Consolas,monospace" font-size="11" font-weight="700" letter-spacing="0.14em" fill="#fff7ed" text-anchor="middle">
-    <g transform="translate(60,80)"><rect width="170" height="80" rx="12" ry="12" fill="#e08a2c" stroke="#9a5416" stroke-width="1.4"/><text x="85" y="36" font-family="Georgia,serif" font-size="18" font-weight="500" fill="#0a0a0b">Capability</text><text x="85" y="60">MODE · CTX · JSON</text></g>
-    <g transform="translate(255,80)"><rect width="170" height="80" rx="12" ry="12" fill="#e08a2c" stroke="#9a5416" stroke-width="1.4"/><text x="85" y="36" font-family="Georgia,serif" font-size="18" font-weight="500" fill="#0a0a0b">Sovereignty</text><text x="85" y="60">RESIDENCY · TIER</text></g>
-    <g transform="translate(450,80)"><rect width="170" height="80" rx="12" ry="12" fill="#e08a2c" stroke="#9a5416" stroke-width="1.4"/><text x="85" y="36" font-family="Georgia,serif" font-size="18" font-weight="500" fill="#0a0a0b">License</text><text x="85" y="60">ALLOWLIST</text></g>
-    <g transform="translate(645,80)"><rect width="170" height="80" rx="12" ry="12" fill="#e08a2c" stroke="#9a5416" stroke-width="1.4"/><text x="85" y="36" font-family="Georgia,serif" font-size="18" font-weight="500" fill="#0a0a0b">Runtime</text><text x="85" y="60">LLAMA.CPP · HF</text></g>
-    <g transform="translate(840,80)"><rect width="170" height="80" rx="12" ry="12" fill="#e08a2c" stroke="#9a5416" stroke-width="1.4"/><text x="85" y="36" font-family="Georgia,serif" font-size="18" font-weight="500" fill="#0a0a0b">Score</text><text x="85" y="60">QUALITY · COST</text></g>
-    <g transform="translate(1035,80)"><rect width="170" height="80" rx="12" ry="12" fill="#0a0a0b" stroke="#0a0a0b" stroke-width="1.4"/><text x="85" y="36" font-family="Georgia,serif" font-size="18" font-weight="500" fill="#fffefa">Audit</text><text x="85" y="60" fill="#e08a2c">DECISION RECORD</text></g>
-  </g>
-  <g stroke="#e08a2c" stroke-width="2" fill="none">
-    <path d="M232 120 L253 120" marker-end="url(#ar2)"/>
-    <path d="M427 120 L448 120" marker-end="url(#ar2)"/>
-    <path d="M622 120 L643 120" marker-end="url(#ar2)"/>
-    <path d="M817 120 L838 120" marker-end="url(#ar2)"/>
-    <path d="M1012 120 L1033 120" marker-end="url(#ar2)"/>
-  </g>
-</svg>"""
+HERO_HTML = """
+<div class="lx-hero">
+  <h1 style="margin:0 0 6px 0;">🧭 MaramaRoute — sovereign router</h1>
+  <p style="margin:0; color:var(--lx-muted);"><em>Ko te marama te tuapapa.</em> Pick the right model from the live LumynaX family of 98+ models — gated on residency, modality, tools, and sovereignty tier.</p>
+  <p style="margin:8px 0 0 0; font-size:0.9em;">Gates: <b>Capability</b> → <b>Sovereignty</b> → <b>License</b> → <b>Runtime</b> → <b>Score</b> → <b>Audit</b></p>
+</div>
+"""
 
 
-BRAND_CSS = """
-:root { --lx-ink:#0a0a0b; --lx-paper:#fffefa; --lx-soft:#f6f0e8; --lx-accent:#e08a2c; --lx-accent-dark:#9a5416; --lx-muted:#726b62; --lx-line:rgba(10,10,11,0.12); }
-body, .gradio-container { background: var(--lx-paper) !important; color: var(--lx-ink) !important; font-family: Aptos, "Avenir Next", "Segoe UI", Helvetica, Arial, sans-serif !important; }
-.lx-shell { width: min(1280px, calc(100% - 48px)); margin: 0 auto; padding-bottom: 60px; }
-.lx-hero { position: relative; padding: 56px 0 28px; border-bottom: 1px solid var(--lx-line); }
-.lx-hero::before { content: ""; position: absolute; top: 0; right: 0; width: min(420px, 42vw); height: 3px; background: var(--lx-accent); }
-.lx-eyebrow { color: var(--lx-accent-dark); font: 700 12px/1.3 ui-monospace, SFMono-Regular, Menlo, Consolas, monospace; letter-spacing: 0.18em; text-transform: uppercase; }
-.lx-hero h1 { margin: 14px 0 12px; font-family: Georgia, Cambria, "Times New Roman", serif; font-size: clamp(40px, 6vw, 80px); line-height: 0.95; font-weight: 500; }
-.lx-hero p.lead { color: var(--lx-muted); max-width: 820px; font-size: clamp(15px, 1.6vw, 19px); line-height: 1.55; }
-.lx-tagline { font-family: Georgia, Cambria, serif; font-style: italic; color: var(--lx-accent-dark); margin-top: 8px; }
-.lx-chips { margin-top: 22px; display: flex; flex-wrap: wrap; gap: 10px; }
-.lx-chips span { border: 1px solid var(--lx-line); border-radius: 999px; padding: 8px 12px; background: #fff; color: var(--lx-muted); font: 700 11px/1 ui-monospace, SFMono-Regular, Menlo, Consolas, monospace; letter-spacing: 0.08em; text-transform: uppercase; }
-.lx-explainer { background: var(--lx-soft); border: 1px solid var(--lx-line); border-left: 4px solid var(--lx-accent); border-radius: 10px; padding: 20px 24px; margin: 24px 0; }
-.lx-explainer h3 { margin: 0 0 8px; font-family: Georgia, Cambria, serif; font-size: 22px; font-weight: 500; }
-.lx-result-allow { background: #f0f7ec; border: 1px solid #4d6b44; border-left: 6px solid #4d6b44; padding: 20px; border-radius: 10px; }
-.lx-result-deny { background: #fbeded; border: 1px solid #b03a3a; border-left: 6px solid #b03a3a; padding: 20px; border-radius: 10px; }
-.gradio-container button.primary { background: var(--lx-ink) !important; border-color: var(--lx-ink) !important; color: #fff !important; border-radius: 999px !important; font-weight: 700 !important; }
-.gradio-container button.primary:hover { background: var(--lx-accent-dark) !important; border-color: var(--lx-accent-dark) !important; }
-.gradio-container button:not(.primary) { background: #fff !important; border-color: var(--lx-line) !important; color: var(--lx-ink) !important; border-radius: 999px !important; }
-.gradio-container textarea, .gradio-container input, .gradio-container .code { background: #fff !important; color: var(--lx-ink) !important; border-color: var(--lx-line) !important; border-radius: 12px !important; }
-.gradio-container label, .gradio-container .block-title, .gradio-container .block-label { color: var(--lx-accent-dark) !important; font: 700 11px/1.2 ui-monospace, SFMono-Regular, Menlo, Consolas, monospace !important; letter-spacing: 0.12em !important; text-transform: uppercase !important; }
-footer { display: none !important; }
-"""
+@dataclass
+class RouteDecision:
+    pick: Optional[dict] = None
+    score: float = 0.0
+    runners_up: list = field(default_factory=list)
+    rejected: list = field(default_factory=list)
+    reason: str = ""
 
 
-def run(req_text: str) -> tuple[str, str, str, str]:
+def _load_registry() -> dict[str, Any]:
     try:
-        request = json.loads(req_text)
-    except Exception as exc:
-        err = f'<div class="lx-result-deny"><h3>Parse error</h3><pre>{type(exc).__name__}: {exc}</pre></div>'
-        return err, "", "", ""
-    decision = route(REGISTRY, request)
-    if decision.selected:
-        s = decision.selected
-        rows = "".join(
-            f"<tr><td><b>{k}</b></td><td><code>{v}</code></td></tr>" for k, v in [
-                ("Repo", s["repo_id"]), ("Family", s.get("family", "—")),
-                ("Runtime", s.get("runtime", "—")), ("Modalities", ", ".join(s.get("modalities", []))),
-                ("Context", f"{s.get('context_tokens', '—')} tok"),
-                ("Sovereignty tier", s.get("sovereignty_tier", "—")),
-                ("Quality / cost rank", f"{s.get('quality_rank', '—')} / {s.get('cost_rank', '—')}"),
-                ("Tools / JSON", f"{s.get('supports_tools')} / {s.get('supports_json')}"),
-                ("Score", s.get("_score", "—")),
-            ]
-        )
-        fb_rows = "".join(f'<li><code>{fb["repo_id"]}</code> · score <b>{fb["_score"]}</b></li>' for fb in decision.fallbacks) or "<li>—</li>"
-        summary = (
-            f'<div class="lx-result-allow">'
-            f'<h3>✓ Selected — <code>{s["repo_id"]}</code></h3>'
-            f'<table style="margin-top:8px"><tbody>{rows}</tbody></table>'
-            f'<p style="margin-top:14px"><b>Fallbacks (next-best matches)</b></p><ol>{fb_rows}</ol>'
-            f'</div>'
-        )
-        explainer = (
-            "### Why this model won\n\n"
-            f"The router scores each candidate on **quality** (lower rank = better, weighted 2&times;), **sovereignty tier** (weighted 1.5&times;), "
-            f"and **cost-fit** (lower = lighter, weighted 0.5&times;). Matching the **task tag** (`{request.get('task_type', '—')}`) adds +3. "
-            f"`requires_local` with a `llama_cpp` runtime adds +1. Frontier-size models (>50B active) lose 0.5.\n\n"
-            f"**This request asked for:** `{request.get('task_type')}` · modalities `{request.get('modalities')}` · "
-            f"jurisdiction `{request.get('jurisdiction')}` · sensitivity `{request.get('data_sensitivity')}` · "
-            f"local `{request.get('requires_local')}` · context ≥ `{request.get('min_context_tokens')}`.\n\n"
-            f"**Rejected candidates:** {len(decision.rejected)}. Top reasons are jurisdiction mismatch, missing modality, runtime not local, or sovereignty tier too low for sensitive data."
-        )
+        p = hf_hub_download(repo_id=REGISTRY_REPO, filename=REGISTRY_PATH,
+                            repo_type="model", token=os.environ.get("HF_TOKEN"))
+        return json.loads(open(p, encoding="utf-8").read())
+    except Exception as e:
+        return {"registry_id": "fallback-empty", "models": [], "_error": str(e)}
+
+
+REGISTRY = _load_registry()
+
+
+def route(prompt: str, modalities, requires_local: bool, requires_tools: bool,
+          requires_json: bool, jurisdiction: str, max_params_b: float, task_hint: str) -> RouteDecision:
+    models = REGISTRY.get("models", [])
+    d = RouteDecision()
+    candidates = []
+    for m in models:
+        if any(mod not in (m.get("modalities") or []) for mod in modalities):
+            d.rejected.append((m["repo_id"], "modality")); continue
+        if requires_local and (m.get("sovereignty_tier") or 5) < 3:
+            d.rejected.append((m["repo_id"], "sovereignty<3")); continue
+        if jurisdiction and jurisdiction not in (m.get("residency") or []):
+            d.rejected.append((m["repo_id"], f"residency!={jurisdiction}")); continue
+        if requires_tools and not m.get("supports_tools"):
+            d.rejected.append((m["repo_id"], "no tools")); continue
+        if requires_json and not m.get("supports_json"):
+            d.rejected.append((m["repo_id"], "no json")); continue
+        if max_params_b > 0 and (m.get("total_params_b") or 0) > max_params_b:
+            d.rejected.append((m["repo_id"], f"params>{max_params_b}")); continue
+        q = int(m.get("quality_rank") or 5)
+        s = int(m.get("sovereignty_tier") or 3)
+        c = int(m.get("cost_rank") or 5)
+        score = (6 - q) * 2 + s * 1.5 + (6 - c) * 0.5
+        if task_hint:
+            tags = " ".join(m.get("tags") or []).lower() + " " + m["model_id"].lower()
+            if task_hint.lower() in tags:
+                score += 3
+        if max_params_b > 0 and (m.get("total_params_b") or 0) <= max_params_b * 0.5:
+            score += 1
+        candidates.append((score, m))
+    candidates.sort(key=lambda x: -x[0])
+    if candidates:
+        d.score, d.pick = candidates[0]
+        d.runners_up = candidates[1:5]
+        d.reason = f"top-of-{len(candidates)} candidates after 5 gates"
     else:
-        summary = (
-            f'<div class="lx-result-deny"><h3>✗ No model satisfies the request</h3>'
-            f'<p>{len(decision.rejected)} candidates were rejected. See "Rejected candidates" for the gate each one failed.</p></div>'
-        )
-        explainer = "### Nothing matched\n\nRelax one constraint and try again: drop `requires_local`, broaden `modalities`, lower `min_context_tokens`, or change `data_sensitivity`."
-
-    selected_json = json.dumps(decision.selected, indent=2) if decision.selected else "{}"
-    rejected_json = json.dumps(decision.rejected[:10] + ([{"note": f"... {len(decision.rejected) - 10} more"}] if len(decision.rejected) > 10 else []), indent=2)
-    return summary, explainer, selected_json, rejected_json
-
-
-with gr.Blocks(theme=gr.themes.Soft(primary_hue="orange", neutral_hue="stone"), css=BRAND_CSS, title="MaramaRoute Live") as demo:
-    with gr.Column(elem_classes="lx-shell"):
-        gr.HTML(
-            f"""
-            <section class="lx-hero">
-              <div class="lx-eyebrow">AbteeX AI Labs · Aotearoa New Zealand · Routing across {MODEL_COUNT} LumynaX models</div>
-              <h1>MaramaRoute <span style="color:#e08a2c">Live</span></h1>
-              <p class="lead">A <b>sovereign model router</b> for the LumynaX release family. Paste a request &mdash; modality, jurisdiction, data sensitivity, runtime constraints &mdash; and see which model wins, with full fallback chain and rejection reasons for every other candidate.</p>
-              <p class="lx-tagline">"Ko te mārama te tūāpapa." — the light is the foundation.</p>
-              <div class="lx-chips">
-                <span>{MODEL_COUNT} models</span><span>Six gates</span><span>Deterministic</span><span>Sovereignty-weighted</span><span>Aotearoa NZ kaupapa</span>
-              </div>
-            </section>
-            """
-        )
-
-        gr.HTML(f'<div style="margin: 20px 0 8px"><div style="font: 700 11px/1.2 ui-monospace, Menlo, Consolas, monospace; letter-spacing: 0.18em; color:#9a5416; text-transform:uppercase">Architecture · Six gates</div>{HERO_SVG}</div>')
-
-        gr.HTML(
-            """
-            <div class="lx-explainer">
-              <h3>How the router decides</h3>
-              <ol>
-                <li><b>Capability gate</b> — model must support the request's modalities, context window, tool calling, and JSON mode.</li>
-                <li><b>Sovereignty gate</b> — request's jurisdiction must be in the model's residency; sensitive data requires sovereignty tier ≥ 2.</li>
-                <li><b>License gate</b> — optional license allowlist and model-card provenance.</li>
-                <li><b>Runtime gate</b> — <code>requires_local</code> excludes hosted-only runtimes.</li>
-                <li><b>Score</b> — candidates are scored on quality, cost-fit, sovereignty tier, and task-tag match.</li>
-                <li><b>Audit</b> — decision, selected model, fallbacks, and rejection reasons are persisted.</li>
-              </ol>
-            </div>
-            """
-        )
-
-        with gr.Row():
-            with gr.Column():
-                request_box = gr.Code(value=json.dumps(EXAMPLE_CODE, indent=2), language="json", label="Request (JSON)", lines=20)
-                with gr.Row():
-                    route_btn = gr.Button("Route", variant="primary")
-                    ex_code = gr.Button("Restricted code")
-                    ex_reason = gr.Button("Reasoning + tools")
-                    ex_mm = gr.Button("Public multimodal")
-                    ex_emb = gr.Button("Restricted embedding")
-                    ex_tiny = gr.Button("Tiny / smoke")
-            with gr.Column():
-                summary = gr.HTML(value='<div style="color:#726b62; padding:18px"><em>Press <b>Route</b> to evaluate a request against the registry.</em></div>')
-        explainer = gr.Markdown()
-        with gr.Row():
-            sel_json = gr.Code(language="json", label="Selected model (full record)", lines=18)
-            rej_json = gr.Code(language="json", label="Rejected candidates (top 10)", lines=18)
+        d.reason = "no candidates survived gating"
+    return d
+
+
+def render(prompt: str, modalities, requires_local: bool, requires_tools: bool,
+           requires_json: bool, jurisdiction: str, max_params_b: float, task_hint: str) -> str:
+    d = route(prompt, modalities, requires_local, requires_tools, requires_json,
+              jurisdiction, max_params_b, task_hint)
+    if not d.pick:
+        return f'<div class="lx-pick" style="background:#fde8e8;border-left-color:#c1351a;"><b>No candidate matches your filters.</b><br/>{d.reason}<br/>Tried {len(d.rejected)} models, all rejected.</div>'
+    p = d.pick
+    runners_html = "<table class='lx-runners'><tr><th>Runner-up</th><th>Score</th><th>Params</th></tr>"
+    for s, m in d.runners_up:
+        tp = m.get("total_params_b"); ap = m.get("active_params_b")
+        runners_html += f"<tr><td><code>{m['repo_id'].split('/')[-1]}</code></td><td>{s:.2f}</td><td>{tp}B{f'/{ap}Ba' if ap else ''}</td></tr>"
+    runners_html += "</table>"
+    tp = p.get("total_params_b"); ap = p.get("active_params_b")
+    p_str = f"{tp}B" + (f" / {ap}B active" if ap else "")
+    return f"""
+<div class='lx-pick'>
+  <div style='display:flex; justify-content:space-between; align-items:baseline;'>
+    <h2 style='margin:0;'>🎯 {p['title']}</h2>
+    <span style='font-family:monospace;'>score {d.score:.2f}</span>
+  </div>
+  <p style='margin:6px 0 10px 0;'><a href='https://huggingface.co/{p['repo_id']}' target='_blank'><code>{p['repo_id']}</code></a></p>
+  <div style='display:grid; grid-template-columns:1fr 1fr 1fr; gap:8px; font-size:0.95em;'>
+    <div><b>Params:</b> {p_str}</div>
+    <div><b>Context:</b> {p.get('context_tokens','—')}</div>
+    <div><b>Runtime:</b> {p.get('runtime','—')}</div>
+    <div><b>Modalities:</b> {', '.join(p.get('modalities') or [])}</div>
+    <div><b>Sovereignty:</b> tier {p.get('sovereignty_tier')}</div>
+    <div><b>License:</b> {p.get('license_id','—')}</div>
+  </div>
+  <p style='margin:10px 0 4px 0; font-size:0.9em; color:var(--lx-muted);'>{d.reason}. Run with <code>hf download {p['repo_id']}</code> or <code>lumynax run {p['repo_id'].split('/')[-1]}</code>.</p>
+</div>
+<div class='lx-runner-up'>
+  <b>Runner-ups (next 4):</b>
+  {runners_html}
+</div>
+"""
 
-        route_btn.click(run, inputs=request_box, outputs=[summary, explainer, sel_json, rej_json])
-        ex_code.click(lambda: json.dumps(EXAMPLE_CODE, indent=2), outputs=request_box)
-        ex_reason.click(lambda: json.dumps(EXAMPLE_REASON, indent=2), outputs=request_box)
-        ex_mm.click(lambda: json.dumps(EXAMPLE_MULTIMODAL, indent=2), outputs=request_box)
-        ex_emb.click(lambda: json.dumps(EXAMPLE_EMBEDDING, indent=2), outputs=request_box)
-        ex_tiny.click(lambda: json.dumps(EXAMPLE_TINY, indent=2), outputs=request_box)
 
-        gr.HTML(
-            """
-            <div style="margin-top:32px; padding-top:24px; border-top:1px solid rgba(10,10,11,0.12); text-align:center; color:#726b62; font-size:13px">
-              <em>Local roots, global work. · Sovereignty is a design property, not a deployment option.</em><br/>
-              <b><a href="https://huggingface.co/AbteeXAILab/marama-route" style="color:#9a5416">Model repo</a></b> ·
-              <b><a href="https://huggingface.co/AbteeXAILab/sovereigncode" style="color:#9a5416">SovereignCode</a></b> ·
-              <b><a href="https://huggingface.co/spaces/AbteeXAILab/lumynax-live-demo" style="color:#9a5416">Live demo</a></b> ·
-              <b><a href="https://abteex.com" style="color:#9a5416">abteex.com</a></b> ·
-              <b><a href="https://lumynax.com" style="color:#9a5416">lumynax.com</a></b>
-            </div>
-            """
-        )
+REGISTRY_NOTE = f"Loaded {len(REGISTRY.get('models', []))} models from {REGISTRY_REPO}"
+
+with gr.Blocks(css=BRAND_CSS, title="LumynaX · MaramaRoute") as demo:
+    gr.HTML(HERO_HTML)
+    gr.Markdown(f"_{REGISTRY_NOTE}_")
+    with gr.Row():
+        with gr.Column(scale=2):
+            prompt = gr.Textbox(label="Prompt or task", placeholder="What do you need the model for?", lines=2)
+            with gr.Row():
+                modalities = gr.CheckboxGroup(["text", "vision", "audio"], value=["text"], label="Required modalities")
+                jurisdiction = gr.Dropdown(["NZ", "AU", "global", ""], value="NZ", label="Jurisdiction (residency)")
+            with gr.Row():
+                requires_local = gr.Checkbox(False, label="Requires sovereignty tier ≥ 3 (local-runnable)")
+                requires_tools = gr.Checkbox(False, label="Requires tool-calling")
+                requires_json = gr.Checkbox(False, label="Requires JSON-mode")
+            with gr.Row():
+                max_params_b = gr.Slider(0, 700, value=0, step=1, label="Max total params (B). 0 = unlimited")
+                task_hint = gr.Textbox(label="Task hint (matches tags + slug)", placeholder="e.g. coder, reasoning, vision, translate")
+            go = gr.Button("Route", variant="primary")
+        with gr.Column(scale=3):
+            out = gr.HTML()
+    gr.Examples(
+        examples=[
+            ["Fix this Python bug", ["text"], True, False, False, "NZ", 100, "coder"],
+            ["Describe what's in this image", ["text", "vision"], False, False, False, "NZ", 0, "vision"],
+            ["Solve this proof", ["text"], False, False, False, "NZ", 0, "reasoning"],
+            ["Embed these documents", ["text"], True, False, True, "NZ", 5, "embedding"],
+            ["Translate to te reo Maori", ["text"], True, False, False, "NZ", 5, "translate"],
+            ["Long-context document analysis (500K tokens)", ["text"], False, False, False, "NZ", 50, "long-context"],
+        ],
+        inputs=[prompt, modalities, requires_local, requires_tools, requires_json, jurisdiction, max_params_b, task_hint],
+    )
+    go.click(render, [prompt, modalities, requires_local, requires_tools, requires_json, jurisdiction, max_params_b, task_hint], out)
+    gr.Markdown("---\n*Real router · live registry · 98 models · Ko te marama te tuapapa.*")
 
 
 if __name__ == "__main__":

requirements.txt

@@ -1,2 +1,2 @@
-gradio>=5.50,<6
-huggingface-hub>=0.25
+gradio==5.50.0
+huggingface_hub>=0.27