feat(provider): add lazy Featherless model discovery

2026-10-06 06:52:20 +02:00 · 2026-09-22 13:12:19 +01:00
parent 2b2bf0fb90
commit 8bb48780a1
6 changed files with 1194 additions and 15 deletions
@@ -9,6 +9,7 @@ import ipaddress
 import socket
 import time as _time
 import logging
+import threading
 import httpx
 from datetime import datetime
 from typing import List, Dict, Any, Optional
@@ -17,6 +18,9 @@ from fastapi import APIRouter, HTTPException, Form, Query, Body, Request, Respon
 from pydantic import BaseModel
 from fastapi.responses import StreamingResponse
 from core.database import SessionLocal, ModelEndpoint, Session as DbSession
+
+_featherless_search_cache: Dict[tuple, tuple[float, Dict[str, Any]]] = {}
+_featherless_search_cache_lock = threading.Lock()
 try:
    from core.log_safety import redact_url as _redact_url_for_log
 except ModuleNotFoundError:
@@ -854,6 +858,8 @@ def _effective_endpoint_kind(ep: Any, base_url: str) -> str:
    kind = _endpoint_kind(ep)
    if kind != "auto":
        return kind
+    if _host_match(base_url, "featherless.ai"):
+        return "api"
    if getattr(ep, "api_key", None) and not _is_ollama_base(base_url):
        try:
            path = (urlparse(base_url).path or "").rstrip("/")
@@ -1015,6 +1021,8 @@ def _probe_endpoint(base_url: str, api_key: str = None, timeout: int = 5) -> Lis
        if api_key:
            return fetch_available_models(api_key, timeout=timeout)
        return []
+    if provider == "featherless" or _host_match(base, "featherless.ai"):
+        return []
    if _is_google_api_base(base):
        try:
            models = _probe_google_models(base, api_key, timeout=timeout)
@@ -1165,6 +1173,31 @@ def _ping_endpoint(base_url: str, api_key: str = None, timeout: float = 1.5) ->

    last_error: Optional[str] = None

+    if _host_match(base, "featherless.ai") or _safe_detect_provider(base) == "featherless":
+        plan_base = base if base.endswith("/v1") else f"{base}/v1"
+        plan_url = f"{plan_base}/plan"
+        try:
+            r = httpx.get(plan_url, headers=headers, timeout=timeout, verify=llm_verify())
+            result = _result_from_response(r)
+            if result["reachable"]:
+                return result
+            if r.status_code in (401, 403):
+                return {"reachable": False, "status_code": r.status_code, "error": "Featherless API key invalid or unauthorized"}
+        except Exception as e:
+            last_error = str(e)[:120]
+
+        try:
+            models_url = f"{plan_base}/models?available_on_current_plan=true&status=active&conversational=true&page=1&per_page=1"
+            r = httpx.get(models_url, headers=headers, timeout=timeout, verify=llm_verify())
+            result = _result_from_response(r)
+            if result["reachable"]:
+                return result
+            if r.status_code in (401, 403):
+                return {"reachable": False, "status_code": r.status_code, "error": "Featherless API key invalid or unauthorized"}
+            return result
+        except Exception as e:
+            return {"reachable": False, "status_code": None, "error": str(e)[:120]}
+
    try:
        if looks_like_ollama:
            root = base
@@ -1517,6 +1550,8 @@ def setup_model_routes(model_discovery):
        }
        if not base:
            return False, info
+        if _host_match(base, "featherless.ai") or _safe_detect_provider(base) == "featherless":
+            return False, info
        if state.get("inflight"):
            return False, info
        if mode in ("manual", "disabled") and not force:
@@ -2046,9 +2081,10 @@ def setup_model_routes(model_discovery):
                if _picker_requires_pinning(base, kind) and pinned and not _has_explicit_pinned_models(r):
                    r.pinned_models = json.dumps(pinned)
                    upgraded_legacy_pins = True
-                model_inventory_count = len(_merge_model_ids(all_models, pinned))
+                is_featherless = _host_match(base, "featherless.ai") or _safe_detect_provider(base) == "featherless"
+                model_inventory_count = len(pinned) if is_featherless else len(_merge_model_ids(all_models, pinned))
                picker_requires_pinning = _picker_requires_pinning(base, kind)
-                status = "online" if (all_models or visible or pinned) else ("empty" if r.is_enabled else "offline")
+                status = "online" if (all_models or visible or pinned or (is_featherless and r.is_enabled)) else ("empty" if r.is_enabled else "offline")
                results.append({
                    "id": r.id,
                    "name": r.name,
@@ -2114,11 +2150,19 @@ def setup_model_routes(model_discovery):
        # keep those container-local when the frontend marks them as such.
        base_url = _rewrite_loopback_for_docker(base_url, container_local=_truthy(container_local))

+        is_featherless = _host_match(base_url, "featherless.ai") or _safe_detect_provider(base_url) == "featherless"
        # Auto-generate name from URL if not provided
        if not name.strip():
-            name = base_url.replace("http://", "").replace("https://", "").split("/")[0]
+            if is_featherless:
+                name = "Featherless.ai"
+            else:
+                name = base_url.replace("http://", "").replace("https://", "").split("/")[0]

        requested_kind = _normalize_endpoint_kind(endpoint_kind)
+        if is_featherless and requested_kind == "auto":
+            requested_kind = "api"
+        if is_featherless and not pinned_models.strip():
+            pinned_models = "[]"
        refresh_mode = _normalize_endpoint_refresh_mode(model_refresh_mode, requested_kind, base_url)
        refresh_interval = _parse_positive_int(model_refresh_interval, minimum=30, maximum=86400)
        refresh_timeout = _parse_positive_int(model_refresh_timeout, minimum=1, maximum=60)
@@ -2212,6 +2256,8 @@ def setup_model_routes(model_discovery):
                existing_models = _cached_model_ids(existing)
                _existing_pinned = _normalize_model_ids(getattr(existing, "pinned_models", None))
                existing_kind = _effective_endpoint_kind(existing, existing.base_url)
+                is_existing_featherless = _host_match(existing.base_url, "featherless.ai") or _safe_detect_provider(existing.base_url) == "featherless"
+                existing_status = "online" if (existing.is_enabled and is_existing_featherless) else ("online" if (existing_models or _existing_pinned) else ("empty" if existing.is_enabled else "offline"))
                return {
                    "id": existing.id,
                    "name": existing.name,
@@ -2224,8 +2270,8 @@ def setup_model_routes(model_discovery):
                        existing.pinned_models,
                    ),
                    "pinned_models": _existing_pinned,
-                    "online": True,
-                    "status": "online",
+                    "online": existing_status != "offline",
+                    "status": existing_status,
                    "existing": True,
                    "endpoint_kind": existing_kind,
                    "category": _classify_endpoint(existing.base_url, existing_kind),
@@ -2237,7 +2283,7 @@ def setup_model_routes(model_discovery):
        ping = {"reachable": False, "error": None}
        if (should_probe or requested_kind in ("api", "proxy")) and not model_ids:
            ping = _ping_endpoint(base_url, api_key.strip() or None, timeout=min(explicit_timeout, 10.0))
-        if require_model_list and not model_ids:
+        if require_model_list and not model_ids and not is_featherless:
            raise HTTPException(400, _model_endpoint_error_message(base_url, ping))

        ep_id = str(uuid.uuid4())[:8]
@@ -2267,8 +2313,8 @@ def setup_model_routes(model_discovery):
                model_refresh_mode=refresh_mode,
                model_refresh_interval=refresh_interval,
                model_refresh_timeout=refresh_timeout,
-                cached_models=json.dumps(model_ids) if model_ids else None,
-                pinned_models=json.dumps(_pinned) if _pinned else None,
+                cached_models=None if is_featherless else (json.dumps(model_ids) if model_ids else None),
+                pinned_models=json.dumps(_pinned) if (is_featherless or _pinned) else None,
                supports_tools=_st,
                owner=_owner_val,
            )
@@ -2308,6 +2354,8 @@ def setup_model_routes(model_discovery):
            db.close()

        # Return immediately — probing happens via the separate /probe SSE endpoint
+        is_online = bool(model_ids) or bool(_pinned) or bool(ping.get("reachable")) or (is_featherless and ping.get("reachable"))
+        is_status = "online" if (model_ids or _pinned or (is_featherless and ping.get("reachable"))) else ("loading" if ping.get("loading") else ("empty" if ping.get("reachable") else "offline"))
        return {
            "id": ep_id,
            "name": name.strip(),
@@ -2316,8 +2364,8 @@ def setup_model_routes(model_discovery):
            "api_key_fingerprint": _api_key_fingerprint(api_key),
            "models": _merge_model_ids(model_ids, _pinned),
            "pinned_models": _pinned,
-            "online": bool(model_ids) or bool(_pinned) or bool(ping.get("reachable")),
-            "status": "online" if (model_ids or _pinned) else ("loading" if ping.get("loading") else ("empty" if ping.get("reachable") else "offline")),
+            "online": is_online,
+            "status": is_status,
            "ping_error": ping.get("error") if ping else None,
            "endpoint_kind": requested_kind,
            "category": _classify_endpoint(base_url, requested_kind),
@@ -2339,14 +2387,19 @@ def setup_model_routes(model_discovery):
        base_url = resolve_url(base_url)
        base_url = _rewrite_loopback_for_docker(base_url)
        requested_kind = _normalize_endpoint_kind(endpoint_kind)
+        is_featherless = _host_match(base_url, "featherless.ai") or _safe_detect_provider(base_url) == "featherless"
+        if is_featherless and requested_kind == "auto":
+            requested_kind = "api"
        configured_timeout = _parse_positive_int(model_refresh_timeout, minimum=1, maximum=60)
        probe_timeout = _explicit_model_list_timeout(base_url, requested_kind, configured_timeout)
        models = _probe_endpoint(base_url, api_key.strip() or None, timeout=probe_timeout)
        ping = {"reachable": True, "error": None} if models else _ping_endpoint(base_url, api_key.strip() or None, timeout=min(probe_timeout, 10.0))
+        is_online = bool(models) or bool(ping.get("reachable")) or (is_featherless and ping.get("reachable"))
+        is_status = "online" if (models or (is_featherless and ping.get("reachable"))) else ("loading" if ping.get("loading") else ("empty" if ping.get("reachable") else "offline"))
        return {
            "base_url": base_url,
-            "online": bool(models) or bool(ping.get("reachable")),
-            "status": "online" if models else ("loading" if ping.get("loading") else ("empty" if ping.get("reachable") else "offline")),
+            "online": is_online,
+            "status": is_status,
            "ping_error": ping.get("error") if ping else None,
            "models": models,
            "count": len(models),
@@ -2531,6 +2584,149 @@ def setup_model_routes(model_discovery):
        finally:
            db.close()

+    @router.get("/model-endpoints/{ep_id}/catalog-search")
+    async def search_endpoint_catalog(
+        ep_id: str,
+        request: Request,
+        q: str = Query(..., min_length=2, max_length=100),
+        page: int = Query(1, ge=1),
+        per_page: int = Query(50, ge=1, le=100),
+    ):
+        """Search catalog for large-inventory providers like Featherless."""
+        require_admin(request)
+        q_clean = q.strip()
+        if len(q_clean) < 2:
+            raise HTTPException(400, "Search query must be at least 2 characters")
+
+        db = SessionLocal()
+        try:
+            ep = db.query(ModelEndpoint).filter(ModelEndpoint.id == ep_id).first()
+            if not ep or not _chatgpt_endpoint_visible(ep, request):
+                raise HTTPException(404, "Endpoint not found")
+            base = _normalize_base(ep.base_url)
+            is_featherless = _host_match(base, "featherless.ai") or _safe_detect_provider(base) == "featherless"
+            if not is_featherless:
+                raise HTTPException(400, "Catalog search is only supported for Featherless endpoints")
+            api_key = _resolve_probe_key(ep) or (ep.api_key.strip() if getattr(ep, "api_key", None) else None)
+            if not api_key:
+                raise HTTPException(400, "Featherless endpoint has no API key configured")
+        finally:
+            db.close()
+
+        try:
+            page = max(int(page or 1), 1)
+        except Exception:
+            page = 1
+        try:
+            per_page = min(max(int(per_page or 50), 1), 100)
+        except Exception:
+            per_page = 50
+
+        # In-memory cache check
+        cache_key = (ep_id, q_clean.lower(), page, per_page)
+        now = _time.time()
+        with _featherless_search_cache_lock:
+            cached_entry = _featherless_search_cache.get(cache_key)
+            if cached_entry:
+                ts, cached_data = cached_entry
+                if now - ts < 45.0:
+                    return cached_data
+                else:
+                    _featherless_search_cache.pop(cache_key, None)
+
+        # Build upstream URL and params
+        models_url = f"{base}/models" if base.endswith("/v1") else f"{base.rstrip('/')}/v1/models"
+        params = {
+            "q": q_clean,
+            "available_on_current_plan": "true",
+            "status": "active",
+            "conversational": "true",
+            "page": page,
+            "per_page": per_page,
+        }
+        headers = {
+            "Authorization": f"Bearer {api_key}",
+            "Accept": "application/json",
+        }
+
+        try:
+            async with httpx.AsyncClient(timeout=10.0, verify=llm_verify()) as client:
+                r = await client.get(models_url, params=params, headers=headers)
+                if r.status_code in (401, 403):
+                    raise HTTPException(r.status_code, "Featherless API key invalid or unauthorized")
+                if r.status_code == 429:
+                    raise HTTPException(429, "Featherless rate limit exceeded; please try again shortly")
+                if r.status_code >= 500:
+                    raise HTTPException(502, f"Featherless upstream error: HTTP {r.status_code}")
+                if r.status_code >= 400:
+                    raise HTTPException(r.status_code, f"Featherless API error: HTTP {r.status_code}")
+                data = r.json()
+        except httpx.HTTPStatusError as exc:
+            code = exc.response.status_code if exc.response is not None else 502
+            if code in (401, 403):
+                raise HTTPException(code, "Featherless API key invalid or unauthorized")
+            if code == 429:
+                raise HTTPException(429, "Featherless rate limit exceeded; please try again shortly")
+            raise HTTPException(502 if code >= 500 else code, f"Featherless API error: HTTP {code}")
+        except httpx.TimeoutException:
+            raise HTTPException(504, "Featherless search request timed out")
+        except HTTPException:
+            raise
+        except Exception as exc:
+            logger.warning("Featherless catalog search failed: %s", exc)
+            raise HTTPException(502, f"Failed to connect to Featherless: {str(exc)[:120]}")
+
+        raw_items = data.get("data") if isinstance(data, dict) else (data if isinstance(data, list) else [])
+        normalized_items = []
+        for m in (raw_items or []):
+            if not isinstance(m, dict):
+                continue
+            m_id = m.get("id")
+            if not m_id or not isinstance(m_id, str):
+                continue
+            normalized_items.append({
+                "id": m_id,
+                "name": m.get("name") or m_id,
+                "context_length": m.get("context_length"),
+                "max_completion_tokens": m.get("max_completion_tokens"),
+                "is_gated": bool(m.get("is_gated", False)),
+                "available_on_current_plan": bool(m.get("available_on_current_plan", True)),
+            })
+
+        total_val = None
+        if isinstance(data, dict):
+            for k in ("total", "count", "total_count"):
+                v = data.get(k)
+                if isinstance(v, (int, float)) and not isinstance(v, bool) and v >= 0:
+                    total_val = int(v)
+                    break
+
+        if total_val is not None:
+            has_more = (page * per_page) < total_val and len(normalized_items) > 0
+        else:
+            has_more = len(normalized_items) == per_page
+
+        result = {
+            "items": normalized_items,
+            "page": page,
+            "per_page": per_page,
+            "has_more": has_more,
+        }
+        if total_val is not None:
+            result["total"] = total_val
+
+        with _featherless_search_cache_lock:
+            if len(_featherless_search_cache) >= 200:
+                expired_keys = [k for k, (t, _) in _featherless_search_cache.items() if now - t >= 45.0]
+                for k in expired_keys:
+                    _featherless_search_cache.pop(k, None)
+                if len(_featherless_search_cache) >= 200:
+                    oldest_key = min(_featherless_search_cache.keys(), key=lambda k: _featherless_search_cache[k][0])
+                    _featherless_search_cache.pop(oldest_key, None)
+            _featherless_search_cache[cache_key] = (now, result)
+
+        return result
+
    @router.get("/default-chat")
    def get_default_chat(request: Request):
        # SECURITY: resolve the default endpoint + model from the CALLER's
@@ -2844,4 +3040,6 @@ def setup_model_routes(model_discovery):
        _save_settings(settings)
        return {"ok": True, "disabled": body.disabled}

+    router._should_refresh_endpoint = _should_refresh_endpoint
+    router._search_endpoint_catalog = search_endpoint_catalog
    return router
@@ -1099,6 +1099,8 @@ def _detect_provider(url: str) -> str:
    from src.copilot import is_copilot_base
    if is_copilot_base(url):
        return "copilot"
+    if _host_match(url, "featherless.ai"):
+        return "featherless"
    if _host_match(url, "cerebras.ai"):
        return "cerebras"
    if _host_match(url, "mistral.ai"):
@@ -1330,6 +1332,7 @@ def _provider_label(url: str) -> str:
    if is_chatgpt_subscription_base(url): return "ChatGPT Subscription"
    from src.copilot import is_copilot_base
    if is_copilot_base(url): return "GitHub Copilot"
+    if _host_match(url, "featherless.ai"): return "Featherless.ai"
    if _host_match(url, "cerebras.ai"):
        return "cerebras"
    if _host_match(url, "mistral.ai"): return "Mistral"
@@ -2314,6 +2314,7 @@
                  <option value="https://api.openai.com/v1" data-logo="openai">OpenAI</option>
                  <option value="copilot" data-logo="github" data-auth-flow="copilot">GitHub Copilot</option>
                  <option value="chatgpt-subscription" data-logo="openai" data-auth-flow="chatgpt-subscription">ChatGPT Subscription</option>
+                  <option value="https://api.featherless.ai/v1" data-logo="featherless">Featherless.ai</option>
                  <option value="https://openrouter.ai/api/v1" data-logo="openrouter">OpenRouter</option>
                  <option value="https://ollama.com/api" data-logo="ollama">Ollama Cloud</option>
                  <option value="https://api.groq.com/openai/v1" data-logo="groq">Groq</option>
@@ -637,6 +637,298 @@ function endpointDetailHtml(ep, category) {
  return `<div class="admin-ep-detail">${parts.join('')}</div>`;
 }

+function renderFeatherlessPanel(panel, ep, row) {
+  const epId = ep.id;
+  const initialPinned = Array.isArray(ep.pinned_models)
+    ? ep.pinned_models
+    : (typeof ep.pinned_models === 'string' ? JSON.parse(ep.pinned_models || '[]') : []);
+  const enabledSet = new Set(initialPinned);
+  const toolModes = typeof ep.model_tool_modes === 'object' && ep.model_tool_modes !== null
+    ? { ...ep.model_tool_modes }
+    : {};
+  panel.dataset.pickerMode = 'pinned';
+
+  panel.innerHTML = `<div class="mcp-tools-header">
+    <span>Featherless Catalog</span>
+  </div>
+  <div class="featherless-panel" style="display:flex;flex-direction:column;gap:12px;padding:6px 0;">
+    <div class="featherless-search-bar" style="position:relative;display:flex;align-items:center;">
+      <input type="search" class="mcp-tools-search featherless-search-input" placeholder="Search Featherless models (min 2 chars)..." style="width:100%;box-sizing:border-box;" data-featherless-search="${esc(epId)}">
+      <span class="featherless-spinner-host" style="display:none;position:absolute;right:8px;font-size:10px;opacity:0.55;">Searching...</span>
+    </div>
+    <div class="featherless-enabled-section">
+      <div style="font-size:11px;font-weight:600;opacity:0.8;margin-bottom:4px;">
+        Enabled models (<span class="featherless-enabled-count">${enabledSet.size}</span>)
+      </div>
+      <div class="featherless-enabled-list mcp-tools-list" style="max-height:160px;overflow-y:auto;"></div>
+    </div>
+    <div class="featherless-results-section">
+      <div style="font-size:11px;font-weight:600;opacity:0.8;margin-bottom:4px;">
+        Search results
+      </div>
+      <div class="featherless-results-list mcp-tools-list" style="max-height:280px;overflow-y:auto;">
+        <span class="featherless-search-hint" style="opacity:0.5;font-size:11px;padding:4px 0;display:block;">Type at least 2 characters to search over 20,000+ models.</span>
+      </div>
+      <div class="featherless-pagination" style="display:none;margin-top:6px;text-align:center;">
+        <button type="button" class="admin-btn-sm featherless-load-more" style="width:100%;">Load more</button>
+      </div>
+    </div>
+  </div>`;
+
+  const searchInput = panel.querySelector('.featherless-search-input');
+  const spinnerHost = panel.querySelector('.featherless-spinner-host');
+  const enabledListEl = panel.querySelector('.featherless-enabled-list');
+  const enabledCountSpan = panel.querySelector('.featherless-enabled-count');
+  const resultsList = panel.querySelector('.featherless-results-list');
+  const paginationHost = panel.querySelector('.featherless-pagination');
+  const loadMoreBtn = panel.querySelector('.featherless-load-more');
+
+  const showSpinner = () => { if (spinnerHost) spinnerHost.style.display = 'inline-flex'; };
+  const hideSpinner = () => { if (spinnerHost) spinnerHost.style.display = 'none'; };
+
+  const formatTokens = (tokens) => {
+    if (!tokens || typeof tokens !== 'number') return '';
+    if (tokens >= 1000000) return `${(tokens / 1000000).toFixed(tokens % 1000000 === 0 ? 0 : 1)}M`;
+    if (tokens >= 1000) return `${Math.round(tokens / 1024)}k`;
+    return String(tokens);
+  };
+
+  const updateHeaderCount = () => {
+    const countBadge = row ? row.querySelector(`[data-adm-ep-models-count="${epId}"]`) : null;
+    if (countBadge) {
+      countBadge.textContent = `${enabledSet.size} models enabled`;
+    }
+    if (enabledCountSpan) {
+      enabledCountSpan.textContent = String(enabledSet.size);
+    }
+    ep.pinned_models = Array.from(enabledSet);
+  };
+
+  const saveState = async () => {
+    try {
+      await fetch(`/api/model-endpoints/${epId}/models`, {
+        method: 'PATCH',
+        headers: { 'Content-Type': 'application/json' },
+        credentials: 'same-origin',
+        body: JSON.stringify({
+          pinned_models: Array.from(enabledSet),
+          model_tool_modes: toolModes,
+        }),
+      });
+      if (typeof _refreshAfterEndpointChange === 'function') {
+        _refreshAfterEndpointChange();
+      }
+    } catch (err) {
+      console.error('Failed to save Featherless model state', err);
+    }
+  };
+
+  const syncSearchCheckboxes = () => {
+    resultsList.querySelectorAll('input[data-featherless-search-id]').forEach(cb => {
+      const id = cb.dataset.featherlessSearchId;
+      cb.checked = enabledSet.has(id);
+    });
+  };
+
+  const renderEnabledList = () => {
+    if (!enabledListEl) return;
+    if (enabledSet.size === 0) {
+      enabledListEl.innerHTML = '<span style="opacity:0.5;font-size:11px;padding:4px 0;display:block;">No models enabled. Search below to add models.</span>';
+      return;
+    }
+    const sortedIds = Array.from(enabledSet).sort((a, b) => a.localeCompare(b));
+    enabledListEl.innerHTML = sortedIds.map(id => {
+      const displayName = id.split('/').pop() || id;
+      const mode = ['none', 'compact', 'full'].includes(String(toolModes[id] || '').toLowerCase())
+        ? String(toolModes[id]).toLowerCase()
+        : '';
+      return `<div title="${esc(id)}" data-ep-model-row data-model-id="${esc(id)}" class="adm-model-row">
+        <label class="adm-model-label">
+          <input type="checkbox" class="adm-cb-hidden" data-featherless-enabled-id="${esc(id)}" checked>
+          <span class="adm-check-dot" aria-hidden="true"></span>
+          <span class="adm-model-name">${esc(displayName)}</span>
+        </label>
+        <div class="adm-model-tools-col">
+          <span class="adm-model-tools-label" title="Select the tool schema profile for this model">Tools</span>
+          <select class="adm-model-tool-mode admin-tools-select" data-ep-model-id="${esc(id)}" data-original-tool-mode="${esc(toolModes[id] || '')}">
+            <option value="" ${mode === '' ? 'selected' : ''}>Auto</option>
+            <option value="full" ${mode === 'full' ? 'selected' : ''}>Regular tools</option>
+            <option value="compact" ${mode === 'compact' ? 'selected' : ''}>Odysseus compact</option>
+            <option value="none" ${mode === 'none' ? 'selected' : ''}>Tools off</option>
+          </select>
+        </div>
+      </div>`;
+    }).join('');
+
+    enabledListEl.querySelectorAll('input[data-featherless-enabled-id]').forEach(cb => {
+      cb.addEventListener('change', () => {
+        const id = cb.dataset.featherlessEnabledId;
+        if (!cb.checked) {
+          enabledSet.delete(id);
+          updateHeaderCount();
+          renderEnabledList();
+          syncSearchCheckboxes();
+          saveState();
+        }
+      });
+    });
+
+    enabledListEl.querySelectorAll('.adm-model-tool-mode').forEach(sel => {
+      sel.addEventListener('change', () => {
+        const id = sel.dataset.epModelId;
+        const val = String(sel.value || '').toLowerCase();
+        if (val) toolModes[id] = val;
+        else delete toolModes[id];
+        saveState();
+      });
+    });
+  };
+
+  renderEnabledList();
+
+  let currentQuery = '';
+  let currentPage = 1;
+  let searchAbortController = null;
+  let searchTimeout = null;
+  let isSearching = false;
+
+  const renderSearchResults = (items, append = false, hasMore = false) => {
+    if (!append) {
+      resultsList.innerHTML = '';
+    }
+    if (!items || items.length === 0) {
+      if (!append) {
+        resultsList.innerHTML = '<span style="opacity:0.5;font-size:11px;padding:4px 0;display:block;">No models found matching your search.</span>';
+      }
+      if (paginationHost) paginationHost.style.display = 'none';
+      return;
+    }
+
+    const itemsHtml = items.map(item => {
+      const isChecked = enabledSet.has(item.id);
+      const displayName = item.name || item.id;
+      const ctx = item.context_length ? `${formatTokens(item.context_length)} ctx` : '';
+      return `<div title="${esc(item.id)}" data-ep-model-row data-model-id="${esc(item.id)}" class="adm-model-row">
+        <label class="adm-model-label" style="width:100%;">
+          <input type="checkbox" class="adm-cb-hidden" data-featherless-search-id="${esc(item.id)}" ${isChecked ? 'checked' : ''}>
+          <span class="adm-check-dot" aria-hidden="true"></span>
+          <span class="adm-model-name" style="flex:1;">${esc(displayName)}</span>
+          ${ctx ? `<span class="admin-badge" style="margin-left:auto;font-size:9px;opacity:0.6;">${esc(ctx)}</span>` : ''}
+        </label>
+      </div>`;
+    }).join('');
+
+    if (append) {
+      resultsList.insertAdjacentHTML('beforeend', itemsHtml);
+    } else {
+      resultsList.innerHTML = itemsHtml;
+    }
+
+    resultsList.querySelectorAll('input[data-featherless-search-id]').forEach(cb => {
+      if (cb.dataset.listenerAttached) return;
+      cb.dataset.listenerAttached = '1';
+      cb.addEventListener('change', () => {
+        const id = cb.dataset.featherlessSearchId;
+        if (cb.checked) {
+          enabledSet.add(id);
+        } else {
+          enabledSet.delete(id);
+        }
+        updateHeaderCount();
+        renderEnabledList();
+        syncSearchCheckboxes();
+        saveState();
+      });
+    });
+
+    if (paginationHost) {
+      paginationHost.style.display = hasMore ? '' : 'none';
+    }
+  };
+
+  if (searchInput) {
+    searchInput.addEventListener('keydown', (e) => {
+      if (e.key === 'Enter') {
+        e.preventDefault();
+        e.stopPropagation();
+      }
+    });
+
+    searchInput.addEventListener('input', () => {
+      if (searchTimeout) clearTimeout(searchTimeout);
+      searchTimeout = setTimeout(async () => {
+        const q = searchInput.value.trim();
+        if (q.length < 2) {
+          if (searchAbortController) searchAbortController.abort();
+          hideSpinner();
+          resultsList.innerHTML = '<span class="featherless-search-hint" style="opacity:0.5;font-size:11px;padding:4px 0;display:block;">Type at least 2 characters to search over 20,000+ models.</span>';
+          if (paginationHost) paginationHost.style.display = 'none';
+          return;
+        }
+
+        if (searchAbortController) {
+          searchAbortController.abort();
+        }
+        searchAbortController = new AbortController();
+        currentQuery = q;
+        currentPage = 1;
+        showSpinner();
+
+        try {
+          const res = await fetch(`/api/model-endpoints/${epId}/catalog-search?q=${encodeURIComponent(q)}&page=1&per_page=50`, {
+            credentials: 'same-origin',
+            signal: searchAbortController.signal,
+          });
+          if (!res.ok) {
+            const errData = await res.json().catch(() => ({}));
+            throw new Error(errData.detail || `HTTP ${res.status}`);
+          }
+          const data = await res.json();
+          renderSearchResults(data.items, false, data.has_more);
+        } catch (err) {
+          if (err.name === 'AbortError') return;
+          resultsList.innerHTML = `<span class="admin-error" style="font-size:11px;padding:4px 0;display:block;">Search failed: ${esc(err.message)}</span>`;
+          if (paginationHost) paginationHost.style.display = 'none';
+        } finally {
+          hideSpinner();
+        }
+      }, 250);
+    });
+  }
+
+  if (loadMoreBtn) {
+    loadMoreBtn.addEventListener('click', async (e) => {
+      e.preventDefault();
+      e.stopPropagation();
+      if (!currentQuery || isSearching) return;
+      isSearching = true;
+      loadMoreBtn.disabled = true;
+      loadMoreBtn.textContent = 'Loading...';
+      currentPage += 1;
+
+      try {
+        const res = await fetch(`/api/model-endpoints/${epId}/catalog-search?q=${encodeURIComponent(currentQuery)}&page=${currentPage}&per_page=50`, {
+          credentials: 'same-origin',
+        });
+        if (!res.ok) {
+          const errData = await res.json().catch(() => ({}));
+          throw new Error(errData.detail || `HTTP ${res.status}`);
+        }
+        const data = await res.json();
+        renderSearchResults(data.items, true, data.has_more);
+      } catch (err) {
+        if (typeof uiModule !== 'undefined' && uiModule?.showToast) {
+          uiModule.showToast(`Failed to load more models: ${err.message}`, 4000);
+        }
+      } finally {
+        isSearching = false;
+        loadMoreBtn.disabled = false;
+        loadMoreBtn.textContent = 'Load more';
+      }
+    });
+  }
+}
+
 // ChatGPT per-endpoint usage panel expanded state persistence.
 // Preserves only endpoint/auth identifiers, never tokens, secrets, or labels.
 const CHATGPT_USAGE_EXPANDED_KEY = 'odysseus-chatgpt-usage-expanded';
@@ -737,7 +1029,7 @@ async function loadEndpoints() {
      const statusBadge = ep.status === 'empty'
        ? '<span class="admin-badge">no models</span>'
        : ep.online
-          ? `<span class="admin-badge">${countText}</span>`
+          ? `<span class="admin-badge" data-adm-ep-models-count="${ep.id}">${countText}</span>`
          : '<span class="admin-badge admin-badge-off">offline</span>';
      const justAddedClass = (_recentlyAddedEpId && String(ep.id) === _recentlyAddedEpId) ? ' adm-ep-just-added' : '';
      const category = ep.category || (_isLocalEndpoint(ep.base_url) ? 'local' : 'api');
@@ -961,7 +1253,7 @@ async function loadEndpoints() {
        // Don't let interactions inside the expanded panel re-fire the
        // expand/collapse handler — the search box was getting closed
        // because clicking it bubbled up to here.
-        if (e.target.closest('.admin-btn-sm, .admin-btn-delete, .mcp-tools-list, .mcp-tools-header, .mcp-tools-search, input, select, label')) return;
+        if (e.target.closest('.admin-btn-sm, .admin-btn-delete, .mcp-tools-list, .mcp-tools-header, .mcp-tools-search, input, select, label, button, .featherless-panel, .featherless-search-bar, [data-ep-model-row]')) return;
        const epId = header.dataset.admEpHeader;
        const panel = row.querySelector(`[data-adm-ep-models-panel="${epId}"]`);
        if (!panel) return;
@@ -974,6 +1266,11 @@ async function loadEndpoints() {
        }
        if (!_modelsLoaded && isOpen) {
          _modelsLoaded = true;
+          const ep = data.find(x => String(x.id) === String(epId));
+          if (ep && isFeatherlessEndpoint(ep)) {
+            renderFeatherlessPanel(panel, ep, row);
+            return;
+          }
          // Our shared whirlpool spinner (consistent with the rest of the app).
          panel.innerHTML = '';
          let _modelsSpin = null;
@@ -1509,6 +1806,12 @@ function initEndpointForm() {
  }

  function _renderEndpointTestResult(msg, res, d) {
+    const isFeatherless = d && d.base_url && /featherless\.ai/i.test(d.base_url);
+    if (res.ok && isFeatherless && d.online) {
+      msg.textContent = 'Online — Featherless.ai catalog ready (search to enable models)';
+      msg.className = 'admin-success';
+      return;
+    }
    if (res.ok && d.status === 'empty') {
      msg.textContent = 'Online — no models found';
      msg.className = 'admin-success';
@@ -1630,9 +1933,13 @@ function initEndpointForm() {
        await loadEndpoints();
        await _selectAddedModelInChat(d);
        const goLink = ' <a href="#" data-go-added-models style="margin-left:6px;text-decoration:underline;color:inherit;font-weight:600;">Added Models →</a>';
+        const isFeatherless = d && d.base_url && /featherless\.ai/i.test(d.base_url);
        if (!d.online) {
          msg.innerHTML = 'Added (endpoint offline — will retry on next load)' + goLink;
          msg.className = 'admin-error';
+        } else if (isFeatherless) {
+          msg.innerHTML = 'Added Featherless.ai — search catalog to enable models' + goLink;
+          msg.className = 'admin-success';
        } else if (d.status === 'empty') {
          msg.innerHTML = 'Added — endpoint reachable, no models found' + goLink;
          msg.className = 'admin-success';
@@ -3809,7 +4116,7 @@ export function close() {
  settingsModule.close();
 }

-export { shouldDisplayEndpointBaseUrl, isFeatherlessEndpoint, endpointDetailHtml };
+export { shouldDisplayEndpointBaseUrl, isFeatherlessEndpoint, endpointDetailHtml, renderFeatherlessPanel };

 const adminModule = { open, close, _initData, get _initialized() { return initialized; } };
 export default adminModule;
@@ -87,6 +87,9 @@ const _PROVIDERS = [
  // NVIDIA / Nemotron (official Simple Icons)
  [/nvidia|nemotron/i,
    '<svg viewBox="0 0 24 24" fill="currentColor"><path d="M8.948 8.798v-1.43a6.7 6.7 0 0 1 .424-.018c3.922-.124 6.493 3.374 6.493 3.374s-2.774 3.851-5.75 3.851c-.398 0-.787-.062-1.158-.185v-4.346c1.528.185 1.837.857 2.747 2.385l2.04-1.714s-1.492-1.952-4-1.952a6.016 6.016 0 0 0-.796.035m0-4.735v2.138l.424-.027c5.45-.185 9.01 4.47 9.01 4.47s-4.08 4.964-8.33 4.964c-.37 0-.733-.035-1.095-.097v1.325c.3.035.61.062.91.062 3.957 0 6.82-2.023 9.593-4.408.459.371 2.34 1.263 2.73 1.652-2.633 2.208-8.772 3.984-12.253 3.984-.335 0-.653-.018-.971-.053v1.864H24V4.063zm0 10.326v1.131c-3.657-.654-4.673-4.46-4.673-4.46s1.758-1.944 4.673-2.262v1.237H8.94c-1.528-.186-2.73 1.245-2.73 1.245s.68 2.412 2.739 3.11M2.456 10.9s2.164-3.197 6.5-3.533V6.201C4.153 6.59 0 10.653 0 10.653s2.35 6.802 8.948 7.42v-1.237c-4.84-.6-6.492-5.936-6.492-5.936z"/></svg>'],
+  // Featherless AI (feather icon)
+  [/featherless/i,
+    '<svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M20.24 12.24a6 6 0 0 0-8.49-8.49L5 10.5V19h8.5z"/><line x1="16" y1="8" x2="2" y2="22"/><line x1="17.5" y1="15" x2="9" y2="15"/></svg>'],
 ];

 // Returns an SVG string for the given model ID, or null if no match
@@ -107,6 +110,7 @@ export function providerLogo(modelId) {
 const _ENDPOINT_LABELS = [
  [/(^|\.)githubcopilot\.com$/i, "GitHub Copilot"],
  [/(^|\.)chatgpt\.com$/i, "ChatGPT Subscription"],
+  [/(^|\.)featherless\.ai$/i, "Featherless.ai"],
  [/(^|\.)openrouter\.ai$/i, "OpenRouter"],
  [/(^|\.)anthropic\.com$/i, "Anthropic"],
  [/(^|\.)openai\.com$/i, "OpenAI"],
@@ -0,0 +1,666 @@
+"""Tests for Featherless provider detection, setup, lazy discovery, and catalog search."""
+
+import asyncio
+import json
+import shutil
+import subprocess
+import sys
+import time
+import types
+from pathlib import Path
+from types import SimpleNamespace
+from unittest.mock import AsyncMock, MagicMock, patch
+
+import httpx
+import pytest
+from fastapi import HTTPException
+
+from tests.helpers.import_state import clear_fake_endpoint_resolver_modules, preserve_import_state
+
+with preserve_import_state("core.database", "src.database", "core.session_manager", "routes.model_routes"):
+    clear_fake_endpoint_resolver_modules()
+
+    if "core.database" not in sys.modules:
+        _core_db = types.ModuleType("core.database")
+        for _name in [
+            "SessionLocal", "ModelEndpoint", "Session", "ChatMessage", "Document",
+            "DocumentVersion", "GalleryImage", "GalleryAlbum", "Note",
+            "CalendarCal", "CalendarEvent", "ScheduledTask", "TaskRun",
+            "McpServer", "ProviderAuthSession", "Base",
+        ]:
+            setattr(_core_db, _name, MagicMock())
+        _core_db.utcnow_naive = MagicMock()
+        sys.modules["core.database"] = _core_db
+
+    import routes.model_routes as model_routes
+    import src.llm_core as llm_core
+    from routes.model_routes import (
+        _effective_endpoint_kind,
+        _probe_endpoint,
+        _ping_endpoint,
+        _picker_requires_pinning,
+        _has_explicit_pinned_models,
+        _picker_models_for_endpoint,
+        _featherless_search_cache,
+        _featherless_search_cache_lock,
+    )
+    from src.llm_core import (
+        _detect_provider,
+        _provider_label,
+        _is_self_hosted_openai_compatible,
+    )
+
+_REPO = Path(__file__).resolve().parent.parent
+_ADMIN_JS = _REPO / "static" / "js" / "admin.js"
+_ROUTER = model_routes.setup_model_routes(model_discovery=None)
+_should_refresh_endpoint = _ROUTER._should_refresh_endpoint
+search_endpoint_catalog = _ROUTER._search_endpoint_catalog
+
+
+def _route_endpoint(router, path, method="GET"):
+    for route in router.routes:
+        if getattr(route, "path", "") == path and method in getattr(route, "methods", set()):
+            return route.endpoint
+    raise AssertionError(f"{method} {path} route not found")
+
+
+# ============================================================
+# 1. Provider Detection & Identification
+# ============================================================
+
+def test_featherless_provider_detection():
+    url = "https://api.featherless.ai/v1"
+    assert _detect_provider(url) == "featherless"
+    assert _provider_label(url) == "Featherless.ai"
+    assert _is_self_hosted_openai_compatible(url) is False
+
+    subdomain_url = "https://eu.featherless.ai/v1"
+    assert _detect_provider(subdomain_url) == "featherless"
+    assert _provider_label(subdomain_url) == "Featherless.ai"
+    assert _is_self_hosted_openai_compatible(subdomain_url) is False
+
+
+def test_featherless_endpoint_kind_is_api_not_proxy():
+    ep = SimpleNamespace(endpoint_kind="auto", api_key="sk-test-key")
+    url = "https://api.featherless.ai/v1"
+    # Keyed /v1 URLs normally resolve to 'proxy', but Featherless must resolve to 'api'
+    assert _effective_endpoint_kind(ep, url) == "api"
+
+
+# ============================================================
+# 2. Probing & Setup Validation
+# ============================================================
+
+def test_featherless_probe_endpoint_bypasses_full_catalog():
+    with patch("httpx.get") as mock_get:
+        models = _probe_endpoint("https://api.featherless.ai/v1", api_key="sk-test-key")
+        # Probe must immediately return [] without making any HTTP request to fetch 20k+ models
+        assert models == []
+        mock_get.assert_not_called()
+
+
+def test_featherless_ping_endpoint_plan_success():
+    resp_plan = MagicMock()
+    resp_plan.status_code = 200
+    resp_plan.text = '{"plan": "pro"}'
+
+    with patch("httpx.get", return_value=resp_plan) as mock_get:
+        res = _ping_endpoint("https://api.featherless.ai/v1", api_key="sk-test-key")
+        assert res["reachable"] is True
+        assert res["status_code"] == 200
+        assert res["error"] is None
+        mock_get.assert_called_once()
+        assert "plan" in mock_get.call_args[0][0]
+
+
+def test_featherless_ping_endpoint_plan_fallback_to_models():
+    # If /v1/plan returns 404, fallback to /v1/models with per_page=1
+    resp_404 = MagicMock()
+    resp_404.status_code = 404
+    resp_404.text = "Not found"
+
+    resp_models = MagicMock()
+    resp_models.status_code = 200
+    resp_models.text = '{"data": [{"id": "model1"}]}'
+
+    with patch("httpx.get", side_effect=[resp_404, resp_models]) as mock_get:
+        res = _ping_endpoint("https://api.featherless.ai/v1", api_key="sk-test-key")
+        assert res["reachable"] is True
+        assert res["status_code"] == 200
+        assert mock_get.call_count == 2
+        assert "available_on_current_plan=true" in mock_get.call_args_list[1][0][0]
+        assert "per_page=1" in mock_get.call_args_list[1][0][0]
+
+
+def test_featherless_ping_endpoint_auth_failure():
+    resp_401 = MagicMock()
+    resp_401.status_code = 401
+    resp_401.text = "Unauthorized"
+
+    with patch("httpx.get", return_value=resp_401):
+        res = _ping_endpoint("https://api.featherless.ai/v1", api_key="bad-key")
+        assert res["reachable"] is False
+        assert res["status_code"] == 401
+        assert "Featherless API key invalid or unauthorized" in res["error"]
+
+
+# ============================================================
+# 3. Background Refresh & Catalog Protection
+# ============================================================
+
+def test_featherless_should_refresh_endpoint_returns_false():
+    ep = SimpleNamespace(
+        id="ep-fl",
+        base_url="https://api.featherless.ai/v1",
+        api_key="sk-test",
+        provider_auth_id=None,
+        cached_models=None,
+        pinned_models="[]",
+    )
+    should_refresh, info = _should_refresh_endpoint(ep, time.time())
+    assert should_refresh is False
+    assert info["base"] == "https://api.featherless.ai/v1"
+
+
+# ============================================================
+# 4. Pinning & Chat Picker Isolation
+# ============================================================
+
+def test_featherless_picker_models_initially_empty():
+    url = "https://api.featherless.ai/v1"
+    kind = "api"
+    assert _picker_requires_pinning(url, kind) is True
+
+    ep = SimpleNamespace(
+        base_url=url,
+        endpoint_kind=kind,
+        pinned_models="[]",
+        cached_models=None,
+        hidden_models=None,
+    )
+    assert _has_explicit_pinned_models(ep) is True
+    visible, pinned = _picker_models_for_endpoint(ep, url, kind)
+    # Default state has 0 models enabled
+    assert visible == []
+    assert pinned == []
+
+
+def test_featherless_picker_models_reflects_pinned_only():
+    url = "https://api.featherless.ai/v1"
+    kind = "api"
+    ep = SimpleNamespace(
+        base_url=url,
+        endpoint_kind=kind,
+        pinned_models=json.dumps(["mistralai/Mistral-7B-Instruct-v0.2", "meta-llama/Llama-3-8B-Instruct"]),
+        cached_models=None,
+        hidden_models=None,
+    )
+    visible, pinned = _picker_models_for_endpoint(ep, url, kind)
+    assert visible == ["mistralai/Mistral-7B-Instruct-v0.2", "meta-llama/Llama-3-8B-Instruct"]
+    assert pinned == ["mistralai/Mistral-7B-Instruct-v0.2", "meta-llama/Llama-3-8B-Instruct"]
+
+
+# ============================================================
+# 5. Catalog Search Route
+# ============================================================
+
+@pytest.mark.asyncio
+async def test_featherless_catalog_search_validation():
+    # q < 2 chars raises HTTPException(400)
+    req = MagicMock()
+    with pytest.raises(HTTPException) as exc_info:
+        await search_endpoint_catalog("ep-1", req, q="a")
+    assert exc_info.value.status_code == 400
+    assert "at least 2 characters" in exc_info.value.detail
+
+
+class _FakeQuery:
+    def __init__(self, ep):
+        self.ep = ep
+
+    def filter(self, *args, **kwargs):
+        return self
+
+    def order_by(self, *args, **kwargs):
+        return self
+
+    def all(self):
+        return [self.ep] if self.ep else []
+
+    def first(self):
+        return self.ep
+
+
+class _FakeDb:
+    def __init__(self, ep):
+        self.ep = ep
+
+    def query(self, *args, **kwargs):
+        return _FakeQuery(self.ep)
+
+    def close(self):
+        pass
+
+
+def test_create_featherless_endpoint(monkeypatch):
+    create = _route_endpoint(_ROUTER, "/api/model-endpoints", "POST")
+    added = []
+    class FakeDb:
+        def __init__(self):
+            self.added = added
+        def query(self, *args, **kwargs):
+            return _FakeQuery(None)
+        def add(self, row):
+            self.added.append(row)
+        def commit(self):
+            pass
+        def close(self):
+            pass
+
+    monkeypatch.setattr(model_routes, "SessionLocal", FakeDb)
+    monkeypatch.setattr(model_routes, "require_admin", lambda r: None)
+    monkeypatch.setattr(model_routes, "_ping_endpoint", lambda *a, **kw: {"reachable": True, "error": None})
+    monkeypatch.setattr(model_routes, "_load_settings", lambda: {})
+    monkeypatch.setattr(model_routes, "_save_settings", lambda s: None)
+
+    req = MagicMock()
+    result = create(
+        req,
+        base_url="https://api.featherless.ai/v1",
+        name="",
+        api_key="sk-test",
+        skip_probe="false",
+        require_models="false",
+        model_type="llm",
+        endpoint_kind="auto",
+        model_refresh_mode="",
+        model_refresh_interval="",
+        model_refresh_timeout="",
+        supports_tools="",
+        pinned_models="",
+        container_local="false",
+        shared="true",
+    )
+
+    assert result["name"] == "Featherless.ai"
+    assert result["endpoint_kind"] == "api"
+    assert result["pinned_models"] == []
+    assert result["models"] == []
+    assert result["online"] is True
+    assert result["status"] == "online"
+
+    assert len(added) == 1
+    ep = added[0]
+    assert ep.name == "Featherless.ai"
+    assert ep.endpoint_kind == "api"
+    assert ep.pinned_models == "[]"
+    assert ep.cached_models is None
+
+
+def test_list_featherless_endpoint(monkeypatch):
+    list_ep = _route_endpoint(_ROUTER, "/api/model-endpoints", "GET")
+    ep = SimpleNamespace(
+        id="ep-fl",
+        name="Featherless.ai",
+        base_url="https://api.featherless.ai/v1",
+        api_key="sk-test",
+        is_enabled=True,
+        cached_models=None,
+        pinned_models="[]",
+        hidden_models=None,
+        endpoint_kind="api",
+        model_type="llm",
+        supports_tools=None,
+        model_refresh_mode="auto",
+        model_refresh_interval=None,
+        model_refresh_timeout=None,
+        owner=None,
+        created_at=None,
+        updated_at=None,
+    )
+    class FakeDb:
+        def query(self, *args, **kwargs):
+            m = MagicMock()
+            m.order_by.return_value.all.return_value = [ep]
+            return m
+        def close(self):
+            pass
+
+    monkeypatch.setattr(model_routes, "SessionLocal", FakeDb)
+    monkeypatch.setattr(model_routes, "require_admin", lambda r: None)
+    monkeypatch.setattr(model_routes, "_disable_stale_cookbook_local_endpoints", lambda db: False)
+
+    req = MagicMock()
+    results = list_ep(req)
+    assert len(results) == 1
+    r = results[0]
+    assert r["name"] == "Featherless.ai"
+    assert r["status"] == "online"
+    assert r["online"] is True
+    assert r["model_count"] == 0
+    assert r["models"] == []
+    assert r["pinned_models"] == []
+
+
+@pytest.mark.asyncio
+async def test_featherless_catalog_search_non_featherless_endpoint(monkeypatch):
+    req = MagicMock()
+    ep_mock = SimpleNamespace(
+        id="ep-openai",
+        base_url="https://api.openai.com/v1",
+        api_key="sk-test",
+    )
+    monkeypatch.setattr(model_routes, "require_admin", lambda r: None)
+    monkeypatch.setattr(model_routes, "_chatgpt_endpoint_visible", lambda ep, req: True)
+    monkeypatch.setattr(model_routes, "SessionLocal", lambda: _FakeDb(ep_mock))
+
+    with pytest.raises(HTTPException) as exc_info:
+        await search_endpoint_catalog("ep-openai", req, q="gpt")
+    assert exc_info.value.status_code == 400
+    assert "only supported for Featherless" in exc_info.value.detail
+
+
+@pytest.mark.asyncio
+async def test_featherless_catalog_search_success_and_caching(monkeypatch):
+    req = MagicMock()
+    ep_mock = SimpleNamespace(
+        id="ep-fl",
+        base_url="https://api.featherless.ai/v1",
+        api_key="sk-test-key",
+    )
+
+    upstream_data = {
+        "data": [
+            {
+                "id": "mistralai/Mistral-7B-Instruct-v0.2",
+                "name": "Mistral 7B Instruct v0.2",
+                "context_length": 32768,
+                "max_completion_tokens": 8192,
+                "is_gated": False,
+                "available_on_current_plan": True,
+            },
+            {
+                "id": "meta-llama/Meta-Llama-3-8B-Instruct",
+                "context_length": 8192,
+            },
+        ]
+    }
+
+    mock_resp = MagicMock()
+    mock_resp.status_code = 200
+    mock_resp.json.return_value = upstream_data
+
+    # Clear cache before test
+    with _featherless_search_cache_lock:
+        _featherless_search_cache.clear()
+
+    monkeypatch.setattr(model_routes, "require_admin", lambda r: None)
+    monkeypatch.setattr(model_routes, "_chatgpt_endpoint_visible", lambda ep, req: True)
+    monkeypatch.setattr(model_routes, "SessionLocal", lambda: _FakeDb(ep_mock))
+
+    with patch("httpx.AsyncClient.get", new_callable=AsyncMock, return_value=mock_resp) as mock_async_get:
+        res1 = await search_endpoint_catalog("ep-fl", req, q="mistral", page=1, per_page=50)
+        assert len(res1["items"]) == 2
+        assert res1["items"][0]["id"] == "mistralai/Mistral-7B-Instruct-v0.2"
+        assert res1["items"][0]["context_length"] == 32768
+        assert res1["items"][1]["name"] == "meta-llama/Meta-Llama-3-8B-Instruct"
+        assert res1["page"] == 1
+        assert res1["per_page"] == 50
+        assert mock_async_get.call_count == 1
+
+        # Check upstream call parameters
+        call_kwargs = mock_async_get.call_args[1]
+        assert call_kwargs["params"]["q"] == "mistral"
+        assert "search" not in call_kwargs["params"]
+        assert call_kwargs["params"]["available_on_current_plan"] == "true"
+        assert call_kwargs["params"]["status"] == "active"
+        assert call_kwargs["params"]["conversational"] == "true"
+        assert call_kwargs["headers"]["Authorization"] == "Bearer sk-test-key"
+
+        # Second call with same query should hit in-memory cache without calling upstream
+        res2 = await search_endpoint_catalog("ep-fl", req, q="mistral", page=1, per_page=50)
+        assert res2 == res1
+        assert mock_async_get.call_count == 1  # Not incremented!
+
+
+@pytest.mark.asyncio
+async def test_featherless_catalog_search_exact_upstream_params(monkeypatch):
+    """Proves exact upstream query parameters: q (not search), filters, page, bounded per_page."""
+    req = MagicMock()
+    ep_mock = SimpleNamespace(
+        id="ep-fl",
+        base_url="https://api.featherless.ai/v1",
+        api_key="sk-test-key",
+    )
+
+    with _featherless_search_cache_lock:
+        _featherless_search_cache.clear()
+
+    monkeypatch.setattr(model_routes, "require_admin", lambda r: None)
+    monkeypatch.setattr(model_routes, "_chatgpt_endpoint_visible", lambda ep, req: True)
+    monkeypatch.setattr(model_routes, "SessionLocal", lambda: _FakeDb(ep_mock))
+
+    mock_resp = MagicMock()
+    mock_resp.status_code = 200
+    mock_resp.json.return_value = {"data": [{"id": "deepseek-ai/DeepSeek-V3"}]}
+
+    with patch("httpx.AsyncClient.get", new_callable=AsyncMock, return_value=mock_resp) as mock_async_get:
+        # Standard query
+        await search_endpoint_catalog("ep-fl", req, q="deepseek", page=2, per_page=50)
+        assert mock_async_get.call_count == 1
+        call_kwargs = mock_async_get.call_args[1]
+        params = call_kwargs["params"]
+
+        assert params["q"] == "deepseek"
+        assert "search" not in params
+        assert params["available_on_current_plan"] == "true"
+        assert params["status"] == "active"
+        assert params["conversational"] == "true"
+        assert params["page"] == 2
+        assert params["per_page"] == 50
+
+        # Bounded per_page: upper bound (500 -> 100)
+        await search_endpoint_catalog("ep-fl", req, q="deepseek-high", page=1, per_page=500)
+        params_upper = mock_async_get.call_args[1]["params"]
+        assert params_upper["per_page"] == 100
+
+        # Bounded page and per_page: lower bound (page 0 -> 1, per_page -5 -> 1)
+        await search_endpoint_catalog("ep-fl", req, q="deepseek-low", page=0, per_page=-5)
+        params_lower = mock_async_get.call_args[1]["params"]
+        assert params_lower["page"] == 1
+        assert params_lower["per_page"] == 1
+
+
+@pytest.mark.asyncio
+async def test_featherless_catalog_pagination_defensive_behavior(monkeypatch):
+    """Tests defensive pagination: total/count metadata vs data-only fallback."""
+    req = MagicMock()
+    ep_mock = SimpleNamespace(
+        id="ep-fl",
+        base_url="https://api.featherless.ai/v1",
+        api_key="sk-test-key",
+    )
+
+    with _featherless_search_cache_lock:
+        _featherless_search_cache.clear()
+
+    monkeypatch.setattr(model_routes, "require_admin", lambda r: None)
+    monkeypatch.setattr(model_routes, "_chatgpt_endpoint_visible", lambda ep, req: True)
+    monkeypatch.setattr(model_routes, "SessionLocal", lambda: _FakeDb(ep_mock))
+
+    mock_resp = MagicMock()
+    mock_resp.status_code = 200
+
+    with patch("httpx.AsyncClient.get", new_callable=AsyncMock, return_value=mock_resp) as mock_async_get:
+        # Case 1: Response with total metadata (page 1 * 50 = 50 < 120 => has_more=True)
+        mock_resp.json.return_value = {
+            "data": [{"id": f"model-{i}"} for i in range(50)],
+            "total": 120,
+        }
+        res1 = await search_endpoint_catalog("ep-fl", req, q="query1", page=1, per_page=50)
+        assert res1["has_more"] is True
+        assert res1["total"] == 120
+        assert len(res1["items"]) == 50
+
+        # Case 2: Response with total metadata reached (page 1 * 50 = 50 >= 50 => has_more=False)
+        mock_resp.json.return_value = {
+            "data": [{"id": f"model-{i}"} for i in range(50)],
+            "total": 50,
+        }
+        res2 = await search_endpoint_catalog("ep-fl", req, q="query2", page=1, per_page=50)
+        assert res2["has_more"] is False
+        assert res2["total"] == 50
+
+        # Case 3: Response with count metadata reached (page 2 * 50 = 100 >= 80 => has_more=False)
+        mock_resp.json.return_value = {
+            "data": [{"id": f"model-{i}"} for i in range(30)],
+            "count": 80,
+        }
+        res3 = await search_endpoint_catalog("ep-fl", req, q="query3", page=2, per_page=50)
+        assert res3["has_more"] is False
+        assert res3["total"] == 80
+
+        # Case 4: Response containing ONLY {"data": [...]} with exactly per_page items => conservative has_more=True
+        mock_resp.json.return_value = {
+            "data": [{"id": f"model-{i}"} for i in range(50)],
+        }
+        res4 = await search_endpoint_catalog("ep-fl", req, q="query4", page=1, per_page=50)
+        assert res4["has_more"] is True
+        assert "total" not in res4
+        assert len(res4["items"]) == 50
+
+        # Case 5: Response containing ONLY {"data": [...]} with fewer than per_page items => has_more=False
+        mock_resp.json.return_value = {
+            "data": [{"id": f"model-{i}"} for i in range(49)],
+        }
+        res5 = await search_endpoint_catalog("ep-fl", req, q="query5", page=1, per_page=50)
+        assert res5["has_more"] is False
+        assert "total" not in res5
+        assert len(res5["items"]) == 49
+
+        # Case 6: Response containing ONLY {"data": []} => has_more=False
+        mock_resp.json.return_value = {
+            "data": [],
+        }
+        res6 = await search_endpoint_catalog("ep-fl", req, q="query6", page=1, per_page=50)
+        assert res6["has_more"] is False
+        assert "total" not in res6
+        assert len(res6["items"]) == 0
+
+
+@pytest.mark.asyncio
+async def test_featherless_catalog_search_error_handling(monkeypatch):
+    req = MagicMock()
+    ep_mock = SimpleNamespace(
+        id="ep-fl",
+        base_url="https://api.featherless.ai/v1",
+        api_key="sk-test-key",
+    )
+
+    monkeypatch.setattr(model_routes, "require_admin", lambda r: None)
+    monkeypatch.setattr(model_routes, "_chatgpt_endpoint_visible", lambda ep, req: True)
+    monkeypatch.setattr(model_routes, "SessionLocal", lambda: _FakeDb(ep_mock))
+
+    # 401 Unauthorized
+    resp_401 = MagicMock()
+    resp_401.status_code = 401
+    with patch("httpx.AsyncClient.get", new_callable=AsyncMock, return_value=resp_401):
+        with pytest.raises(HTTPException) as exc_401:
+            await search_endpoint_catalog("ep-fl", req, q="llama")
+        assert exc_401.value.status_code == 401
+        assert "API key invalid" in exc_401.value.detail
+
+    # 429 Rate Limit
+    resp_429 = MagicMock()
+    resp_429.status_code = 429
+    with patch("httpx.AsyncClient.get", new_callable=AsyncMock, return_value=resp_429):
+        with pytest.raises(HTTPException) as exc_429:
+            await search_endpoint_catalog("ep-fl", req, q="llama")
+        assert exc_429.value.status_code == 429
+        assert "rate limit" in exc_429.value.detail.lower()
+
+    # 504 Timeout
+    with patch("httpx.AsyncClient.get", new_callable=AsyncMock, side_effect=httpx.TimeoutException("Timeout")):
+        with pytest.raises(HTTPException) as exc_504:
+            await search_endpoint_catalog("ep-fl", req, q="llama")
+        assert exc_504.value.status_code == 504
+
+
+# ============================================================
+# 6. Frontend JS Tests (Node)
+# ============================================================
+
+@pytest.mark.skipif(not shutil.which("node"), reason="node not on PATH")
+class TestFeatherlessFrontend:
+    def test_featherless_js_panel_and_helpers(self):
+        js = f"""
+          import fs from 'node:fs';
+          import {{ isChatgptSubscriptionEndpoint }} from '{(_REPO / 'static' / 'js' / 'chatgptSubscriptionUsage.js').as_posix()}';
+          const source = fs.readFileSync('{_ADMIN_JS.as_posix()}', 'utf8');
+          const fnStart = source.indexOf('function shouldDisplayEndpointBaseUrl');
+          const fnEnd = source.indexOf('// ChatGPT per-endpoint usage panel', fnStart);
+          const fnCode = source.slice(fnStart, fnEnd);
+          const fns = new Function('isChatgptSubscriptionEndpoint', 'esc',
+            fnCode + '; return {{ shouldDisplayEndpointBaseUrl, isFeatherlessEndpoint, renderFeatherlessPanel }};'
+          )(isChatgptSubscriptionEndpoint, x => String(x));
+
+          const ep = {{
+            id: 'fl-1',
+            base_url: 'https://api.featherless.ai/v1',
+            provider: 'featherless',
+            pinned_models: ['mistralai/Mistral-7B-Instruct-v0.2']
+          }};
+
+          const isFl = fns.isFeatherlessEndpoint(ep);
+          const showUrl = fns.shouldDisplayEndpointBaseUrl(ep);
+
+          // Test renderFeatherlessPanel DOM construction
+          const mockPanel = {{
+            dataset: {{}},
+            innerHTML: '',
+            querySelector: function(sel) {{
+              if (sel === '.featherless-search-input') return {{ addEventListener: () => {{}}, value: '' }};
+              if (sel === '.featherless-enabled-list') return {{ innerHTML: '', querySelectorAll: () => [] }};
+              if (sel === '.featherless-enabled-count') return {{ textContent: '' }};
+              if (sel === '.featherless-results-list') return {{ innerHTML: '', querySelectorAll: () => [] }};
+              if (sel === '.featherless-pagination') return {{ style: {{}} }};
+              if (sel === '.featherless-load-more') return {{ addEventListener: () => {{}} }};
+              if (sel === '.featherless-spinner-host') return {{ style: {{}} }};
+              return null;
+            }},
+            querySelectorAll: function() {{ return []; }}
+          }};
+
+          const mockRow = {{
+            querySelector: function() {{ return {{ textContent: '' }}; }}
+          }};
+
+          fns.renderFeatherlessPanel(mockPanel, ep, mockRow);
+
+          console.log(JSON.stringify({{
+            isFl,
+            showUrl,
+            pickerMode: mockPanel.dataset.pickerMode,
+            hasHeader: mockPanel.innerHTML.includes('Featherless Catalog'),
+            hasSearchBar: mockPanel.innerHTML.includes('featherless-search-bar'),
+            hasEnabledSection: mockPanel.innerHTML.includes('featherless-enabled-section'),
+            hasResultsSection: mockPanel.innerHTML.includes('featherless-results-section'),
+          }}));
+        """
+        proc = subprocess.run(
+            ["node", "--input-type=module"],
+            input=js,
+            capture_output=True,
+            text=True,
+            cwd=str(_REPO),
+            timeout=30,
+        )
+        assert proc.returncode == 0, proc.stderr
+        data = json.loads(proc.stdout.strip())
+        assert data["isFl"] is True
+        assert data["showUrl"] is False
+        assert data["pickerMode"] == "pinned"
+        assert data["hasHeader"] is True
+        assert data["hasSearchBar"] is True
+        assert data["hasEnabledSection"] is True
+        assert data["hasResultsSection"] is True