feat(analytics): integrate Langfuse with Keycloak SSO

Add certifai-langfuse OIDC client to Keycloak realm export and configure the Langfuse Docker service with Keycloak SSO env vars (shared realm, account linking, local auth disabled). Replace the iframe-based analytics page with an informational landing since cross-origin SSO breaks in iframes. Users open Langfuse in a new tab where the active Keycloak session authenticates them transparently. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
feat(developer): replace placeholder letters with Bootstrap icons on agents page
2026-02-25 20:54:47 +01:00 · 2026-02-25 20:37:58 +01:00 · 2026-02-25 20:35:44 +01:00 · 2026-02-25 18:37:33 +01:00 · 2026-02-25 17:49:56 +01:00 · 2026-02-25 10:27:19 +01:00
29 changed files with 500 additions and 1107 deletions
--- a/.env.example
+++ b/.env.example
@@ -34,11 +34,10 @@ MONGODB_DATABASE=certifai
 SEARXNG_URL=http://localhost:8888

 # ---------------------------------------------------------------------------
-# LiteLLM proxy [OPTIONAL - defaults shown]
+# Ollama LLM instance [OPTIONAL - defaults shown]
 # ---------------------------------------------------------------------------
-LITELLM_URL=http://localhost:4000
-LITELLM_MODEL=qwen3-32b
-LITELLM_API_KEY=
+OLLAMA_URL=http://localhost:11434
+OLLAMA_MODEL=llama3.1:8b

 # ---------------------------------------------------------------------------
 # LibreChat (external chat via SSO) [OPTIONAL - default: http://localhost:3080]
@@ -48,7 +47,7 @@ LIBRECHAT_URL=http://localhost:3080
 # ---------------------------------------------------------------------------
 # LLM Providers (comma-separated list) [OPTIONAL]
 # ---------------------------------------------------------------------------
-LLM_PROVIDERS=litellm
+LLM_PROVIDERS=ollama

 # ---------------------------------------------------------------------------
 # SMTP (transactional email) [OPTIONAL]
--- a/.gitea/workflows/ci.yml
+++ b/.gitea/workflows/ci.yml
@@ -121,13 +121,13 @@ jobs:
        if: always()

  # ---------------------------------------------------------------------------
-  # Stage 4: E2E tests (only on main, after deploy)
+  # Stage 2b: E2E tests (only on main / PRs to main, after quality checks)
  # ---------------------------------------------------------------------------
  e2e:
    name: E2E Tests
    runs-on: docker
-    needs: [deploy]
-    if: github.ref == 'refs/heads/main'
+    needs: [fmt, clippy, audit]
+    if: github.ref == 'refs/heads/main' || github.event_name == 'pull_request'
    container:
      image: rust:1.89-bookworm
    # MongoDB and SearXNG can start immediately (no repo files needed).
@@ -259,7 +259,7 @@ jobs:
  deploy:
    name: Deploy
    runs-on: docker
-    needs: [test]
+    needs: [test, e2e]
    if: github.ref == 'refs/heads/main'
    container:
      image: alpine:latest
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -773,7 +773,6 @@ dependencies = [
 "dioxus-sdk",
 "dotenvy",
 "futures",
- "js-sys",
 "maud",
 "mongodb",
 "petname",
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -61,7 +61,6 @@ secrecy = { version = "0.10", default-features = false, optional = true }
 serde_json = { version = "1.0.133", default-features = false }
 maud = { version = "0.27", default-features = false }
 url = { version = "2.5.4", default-features = false, optional = true }
-js-sys = { version = "0.3", optional = true }
 wasm-bindgen = { version = "0.2", optional = true }
 web-sys = { version = "0.3", optional = true, features = [
    "Clipboard",
@@ -92,7 +91,7 @@ bytes = { version = "1", optional = true }

 [features]
 # default = ["web"]
-web = ["dioxus/web", "dep:reqwest", "dep:web-sys", "dep:wasm-bindgen", "dep:js-sys"]
+web = ["dioxus/web", "dep:reqwest", "dep:web-sys", "dep:wasm-bindgen"]
 server = [
    "dioxus/server",
    "dep:axum",
--- a/assets/i18n/de.json
+++ b/assets/i18n/de.json
@@ -58,15 +58,15 @@
    "title": "Dashboard",
    "subtitle": "KI-Nachrichten und Neuigkeiten",
    "topic_placeholder": "Themenname...",
-    "litellm_settings": "LiteLLM-Einstellungen",
-    "settings_hint": "Leer lassen, um LITELLM_URL / LITELLM_MODEL aus .env zu verwenden",
-    "litellm_url": "LiteLLM-URL",
-    "litellm_url_placeholder": "Verwendet LITELLM_URL aus .env",
+    "ollama_settings": "Ollama-Einstellungen",
+    "settings_hint": "Leer lassen, um OLLAMA_URL / OLLAMA_MODEL aus .env zu verwenden",
+    "ollama_url": "Ollama-URL",
+    "ollama_url_placeholder": "Verwendet OLLAMA_URL aus .env",
    "model": "Modell",
-    "model_placeholder": "Verwendet LITELLM_MODEL aus .env",
+    "model_placeholder": "Verwendet OLLAMA_MODEL aus .env",
    "searching": "Suche laeuft...",
    "search_failed": "Suche fehlgeschlagen: {e}",
-    "litellm_status": "LiteLLM-Status",
+    "ollama_status": "Ollama-Status",
    "trending": "Im Trend",
    "recent_searches": "Letzte Suchen"
  },
@@ -144,16 +144,6 @@
    "email_address": "E-Mail-Adresse",
    "email_placeholder": "kollege@firma.de",
    "send_invite": "Einladung senden",
-    "total_spend": "Gesamtausgaben",
-    "total_tokens": "Tokens gesamt",
-    "model_usage": "Nutzung nach Modell",
-    "model": "Modell",
-    "tokens": "Tokens",
-    "spend": "Ausgaben",
-    "usage_unavailable": "Nutzungsdaten nicht verfuegbar",
-    "loading_usage": "Nutzungsdaten werden geladen...",
-    "prompt_tokens": "Prompt-Tokens",
-    "completion_tokens": "Antwort-Tokens",
    "pricing_title": "Preise",
    "pricing_subtitle": "Waehlen Sie den passenden Plan fuer Ihre Organisation"
  },
--- a/assets/i18n/en.json
+++ b/assets/i18n/en.json
@@ -58,15 +58,15 @@
    "title": "Dashboard",
    "subtitle": "AI news and updates",
    "topic_placeholder": "Topic name...",
-    "litellm_settings": "LiteLLM Settings",
-    "settings_hint": "Leave empty to use LITELLM_URL / LITELLM_MODEL from .env",
-    "litellm_url": "LiteLLM URL",
-    "litellm_url_placeholder": "Uses LITELLM_URL from .env",
+    "ollama_settings": "Ollama Settings",
+    "settings_hint": "Leave empty to use OLLAMA_URL / OLLAMA_MODEL from .env",
+    "ollama_url": "Ollama URL",
+    "ollama_url_placeholder": "Uses OLLAMA_URL from .env",
    "model": "Model",
-    "model_placeholder": "Uses LITELLM_MODEL from .env",
+    "model_placeholder": "Uses OLLAMA_MODEL from .env",
    "searching": "Searching...",
    "search_failed": "Search failed: {e}",
-    "litellm_status": "LiteLLM Status",
+    "ollama_status": "Ollama Status",
    "trending": "Trending",
    "recent_searches": "Recent Searches"
  },
@@ -144,16 +144,6 @@
    "email_address": "Email Address",
    "email_placeholder": "colleague@company.com",
    "send_invite": "Send Invite",
-    "total_spend": "Total Spend",
-    "total_tokens": "Total Tokens",
-    "model_usage": "Usage by Model",
-    "model": "Model",
-    "tokens": "Tokens",
-    "spend": "Spend",
-    "usage_unavailable": "Usage data unavailable",
-    "loading_usage": "Loading usage data...",
-    "prompt_tokens": "Prompt Tokens",
-    "completion_tokens": "Completion Tokens",
    "pricing_title": "Pricing",
    "pricing_subtitle": "Choose the plan that fits your organization"
  },
--- a/assets/i18n/es.json
+++ b/assets/i18n/es.json
@@ -58,15 +58,15 @@
    "title": "Panel de control",
    "subtitle": "Noticias y actualizaciones de IA",
    "topic_placeholder": "Nombre del tema...",
-    "litellm_settings": "Configuracion de LiteLLM",
-    "settings_hint": "Dejar vacio para usar LITELLM_URL / LITELLM_MODEL del archivo .env",
-    "litellm_url": "URL de LiteLLM",
-    "litellm_url_placeholder": "Usa LITELLM_URL del archivo .env",
+    "ollama_settings": "Configuracion de Ollama",
+    "settings_hint": "Dejar vacio para usar OLLAMA_URL / OLLAMA_MODEL del archivo .env",
+    "ollama_url": "URL de Ollama",
+    "ollama_url_placeholder": "Usa OLLAMA_URL del archivo .env",
    "model": "Modelo",
-    "model_placeholder": "Usa LITELLM_MODEL del archivo .env",
+    "model_placeholder": "Usa OLLAMA_MODEL del archivo .env",
    "searching": "Buscando...",
    "search_failed": "La busqueda fallo: {e}",
-    "litellm_status": "Estado de LiteLLM",
+    "ollama_status": "Estado de Ollama",
    "trending": "Tendencias",
    "recent_searches": "Busquedas recientes"
  },
@@ -144,16 +144,6 @@
    "email_address": "Direccion de correo electronico",
    "email_placeholder": "colega@empresa.com",
    "send_invite": "Enviar invitacion",
-    "total_spend": "Gasto total",
-    "total_tokens": "Tokens totales",
-    "model_usage": "Uso por modelo",
-    "model": "Modelo",
-    "tokens": "Tokens",
-    "spend": "Gasto",
-    "usage_unavailable": "Datos de uso no disponibles",
-    "loading_usage": "Cargando datos de uso...",
-    "prompt_tokens": "Tokens de entrada",
-    "completion_tokens": "Tokens de respuesta",
    "pricing_title": "Precios",
    "pricing_subtitle": "Elija el plan que se adapte a su organizacion"
  },
--- a/assets/i18n/fr.json
+++ b/assets/i18n/fr.json
@@ -58,15 +58,15 @@
    "title": "Tableau de bord",
    "subtitle": "Actualites et mises a jour IA",
    "topic_placeholder": "Nom du sujet...",
-    "litellm_settings": "Parametres LiteLLM",
-    "settings_hint": "Laissez vide pour utiliser LITELLM_URL / LITELLM_MODEL du fichier .env",
-    "litellm_url": "URL LiteLLM",
-    "litellm_url_placeholder": "Utilise LITELLM_URL du fichier .env",
+    "ollama_settings": "Parametres Ollama",
+    "settings_hint": "Laissez vide pour utiliser OLLAMA_URL / OLLAMA_MODEL du fichier .env",
+    "ollama_url": "URL Ollama",
+    "ollama_url_placeholder": "Utilise OLLAMA_URL du fichier .env",
    "model": "Modele",
-    "model_placeholder": "Utilise LITELLM_MODEL du fichier .env",
+    "model_placeholder": "Utilise OLLAMA_MODEL du fichier .env",
    "searching": "Recherche en cours...",
    "search_failed": "Echec de la recherche : {e}",
-    "litellm_status": "Statut LiteLLM",
+    "ollama_status": "Statut Ollama",
    "trending": "Tendances",
    "recent_searches": "Recherches recentes"
  },
@@ -144,16 +144,6 @@
    "email_address": "Adresse e-mail",
    "email_placeholder": "collegue@entreprise.com",
    "send_invite": "Envoyer l'invitation",
-    "total_spend": "Depenses totales",
-    "total_tokens": "Tokens totaux",
-    "model_usage": "Utilisation par modele",
-    "model": "Modele",
-    "tokens": "Tokens",
-    "spend": "Depenses",
-    "usage_unavailable": "Donnees d'utilisation indisponibles",
-    "loading_usage": "Chargement des donnees d'utilisation...",
-    "prompt_tokens": "Tokens d'entree",
-    "completion_tokens": "Tokens de reponse",
    "pricing_title": "Tarifs",
    "pricing_subtitle": "Choisissez le plan adapte a votre organisation"
  },
--- a/assets/i18n/pt.json
+++ b/assets/i18n/pt.json
@@ -58,15 +58,15 @@
    "title": "Painel",
    "subtitle": "Noticias e atualizacoes de IA",
    "topic_placeholder": "Nome do topico...",
-    "litellm_settings": "Definicoes do LiteLLM",
-    "settings_hint": "Deixe vazio para usar LITELLM_URL / LITELLM_MODEL do .env",
-    "litellm_url": "URL do LiteLLM",
-    "litellm_url_placeholder": "Utiliza LITELLM_URL do .env",
+    "ollama_settings": "Definicoes do Ollama",
+    "settings_hint": "Deixe vazio para usar OLLAMA_URL / OLLAMA_MODEL do .env",
+    "ollama_url": "URL do Ollama",
+    "ollama_url_placeholder": "Utiliza OLLAMA_URL do .env",
    "model": "Modelo",
-    "model_placeholder": "Utiliza LITELLM_MODEL do .env",
+    "model_placeholder": "Utiliza OLLAMA_MODEL do .env",
    "searching": "A pesquisar...",
    "search_failed": "A pesquisa falhou: {e}",
-    "litellm_status": "Estado do LiteLLM",
+    "ollama_status": "Estado do Ollama",
    "trending": "Em destaque",
    "recent_searches": "Pesquisas recentes"
  },
@@ -144,16 +144,6 @@
    "email_address": "Endereco de Email",
    "email_placeholder": "colleague@company.com",
    "send_invite": "Enviar Convite",
-    "total_spend": "Gasto total",
-    "total_tokens": "Tokens totais",
-    "model_usage": "Uso por modelo",
-    "model": "Modelo",
-    "tokens": "Tokens",
-    "spend": "Gasto",
-    "usage_unavailable": "Dados de uso indisponiveis",
-    "loading_usage": "Carregando dados de uso...",
-    "prompt_tokens": "Tokens de entrada",
-    "completion_tokens": "Tokens de resposta",
    "pricing_title": "Precos",
    "pricing_subtitle": "Escolha o plano adequado a sua organizacao"
  },
--- a/assets/tailwind.css
+++ b/assets/tailwind.css
@@ -1,4 +1,4 @@
-/*! tailwindcss v4.2.1 | MIT License | https://tailwindcss.com */
+/*! tailwindcss v4.2.0 | MIT License | https://tailwindcss.com */
@layer properties;
@layer theme, base, components, utilities;
@layer theme {
@@ -162,6 +162,59 @@
  }
 }
@layer utilities {
+  .diff {
+    @layer daisyui.l1.l2.l3 {
+      position: relative;
+      display: grid;
+      width: 100%;
+      overflow: hidden;
+      webkit-user-select: none;
+      user-select: none;
+      grid-template-rows: 1fr 1.8rem 1fr;
+      direction: ltr;
+      container-type: inline-size;
+      grid-template-columns: auto 1fr;
+      &:focus-visible, &:has(.diff-item-1:focus-visible) {
+        outline-style: var(--tw-outline-style);
+        outline-width: 2px;
+        outline-offset: 1px;
+        outline-color: var(--color-base-content);
+      }
+      &:focus-visible {
+        outline-style: var(--tw-outline-style);
+        outline-width: 2px;
+        outline-offset: 1px;
+        outline-color: var(--color-base-content);
+        .diff-resizer {
+          min-width: 95cqi;
+          max-width: 95cqi;
+        }
+      }
+      &:has(.diff-item-1:focus-visible) {
+        outline-style: var(--tw-outline-style);
+        outline-width: 2px;
+        outline-offset: 1px;
+        .diff-resizer {
+          min-width: 5cqi;
+          max-width: 5cqi;
+        }
+      }
+      @supports (-webkit-overflow-scrolling: touch) and (overflow: -webkit-paged-x) {
+        &:focus {
+          .diff-resizer {
+            min-width: 5cqi;
+            max-width: 5cqi;
+          }
+        }
+        &:has(.diff-item-1:focus) {
+          .diff-resizer {
+            min-width: 95cqi;
+            max-width: 95cqi;
+          }
+        }
+      }
+    }
+  }
  .modal {
    @layer daisyui.l1.l2.l3 {
      pointer-events: none;
@@ -1057,98 +1110,31 @@
      }
    }
  }
-  .range {
+  .chat-bubble {
    @layer daisyui.l1.l2.l3 {
-      appearance: none;
-      webkit-appearance: none;
-      --range-thumb: var(--color-base-100);
-      --range-thumb-size: calc(var(--size-selector, 0.25rem) * 6);
-      --range-progress: currentColor;
-      --range-fill: 1;
-      --range-p: 0.25rem;
-      --range-bg: currentColor;
-      @supports (color: color-mix(in lab, red, red)) {
-        --range-bg: color-mix(in oklab, currentColor 10%, #0000);
-      }
-      cursor: pointer;
-      overflow: hidden;
-      background-color: transparent;
-      vertical-align: middle;
-      width: clamp(3rem, 20rem, 100%);
-      --radius-selector-max: calc(
-      var(--radius-selector) + var(--radius-selector) + var(--radius-selector)
-    );
-      border-radius: calc(var(--radius-selector) + min(var(--range-p), var(--radius-selector-max)));
-      border: none;
-      height: var(--range-thumb-size);
-      [dir="rtl"] & {
-        --range-dir: -1;
-      }
-      &:focus {
-        outline: none;
-      }
-      &:focus-visible {
-        outline: 2px solid;
-        outline-offset: 2px;
-      }
-      &::-webkit-slider-runnable-track {
-        width: 100%;
-        background-color: var(--range-bg);
-        border-radius: var(--radius-selector);
-        height: calc(var(--range-thumb-size) * 0.5);
-      }
-      @media (forced-colors: active) {
-        &::-webkit-slider-runnable-track {
-          border: 1px solid;
-        }
-      }
-      @media (forced-colors: active) {
-        &::-moz-range-track {
-          border: 1px solid;
-        }
-      }
-      &::-webkit-slider-thumb {
-        position: relative;
-        box-sizing: border-box;
-        border-radius: calc(var(--radius-selector) + min(var(--range-p), var(--radius-selector-max)));
-        background-color: var(--range-thumb);
-        height: var(--range-thumb-size);
-        width: var(--range-thumb-size);
-        border: var(--range-p) solid;
-        appearance: none;
-        webkit-appearance: none;
-        top: 50%;
-        color: var(--range-progress);
-        transform: translateY(-50%);
-        box-shadow: 0 -1px oklch(0% 0 0 / calc(var(--depth) * 0.1)) inset, 0 8px 0 -4px oklch(100% 0 0 / calc(var(--depth) * 0.1)) inset, 0 1px currentColor, 0 0 0 2rem var(--range-thumb) inset, calc((var(--range-dir, 1) * -100cqw) - (var(--range-dir, 1) * var(--range-thumb-size) / 2)) 0 0 calc(100cqw * var(--range-fill));
-        @supports (color: color-mix(in lab, red, red)) {
-          box-shadow: 0 -1px oklch(0% 0 0 / calc(var(--depth) * 0.1)) inset, 0 8px 0 -4px oklch(100% 0 0 / calc(var(--depth) * 0.1)) inset, 0 1px color-mix(in oklab, currentColor calc(var(--depth) * 10%), #0000), 0 0 0 2rem var(--range-thumb) inset, calc((var(--range-dir, 1) * -100cqw) - (var(--range-dir, 1) * var(--range-thumb-size) / 2)) 0 0 calc(100cqw * var(--range-fill));
-        }
-      }
-      &::-moz-range-track {
-        width: 100%;
-        background-color: var(--range-bg);
-        border-radius: var(--radius-selector);
-        height: calc(var(--range-thumb-size) * 0.5);
-      }
-      &::-moz-range-thumb {
-        position: relative;
-        box-sizing: border-box;
-        border-radius: calc(var(--radius-selector) + min(var(--range-p), var(--radius-selector-max)));
-        background-color: currentColor;
-        height: var(--range-thumb-size);
-        width: var(--range-thumb-size);
-        border: var(--range-p) solid;
-        top: 50%;
-        color: var(--range-progress);
-        box-shadow: 0 -1px oklch(0% 0 0 / calc(var(--depth) * 0.1)) inset, 0 8px 0 -4px oklch(100% 0 0 / calc(var(--depth) * 0.1)) inset, 0 1px currentColor, 0 0 0 2rem var(--range-thumb) inset, calc((var(--range-dir, 1) * -100cqw) - (var(--range-dir, 1) * var(--range-thumb-size) / 2)) 0 0 calc(100cqw * var(--range-fill));
-        @supports (color: color-mix(in lab, red, red)) {
-          box-shadow: 0 -1px oklch(0% 0 0 / calc(var(--depth) * 0.1)) inset, 0 8px 0 -4px oklch(100% 0 0 / calc(var(--depth) * 0.1)) inset, 0 1px color-mix(in oklab, currentColor calc(var(--depth) * 10%), #0000), 0 0 0 2rem var(--range-thumb) inset, calc((var(--range-dir, 1) * -100cqw) - (var(--range-dir, 1) * var(--range-thumb-size) / 2)) 0 0 calc(100cqw * var(--range-fill));
-        }
-      }
-      &:disabled {
-        cursor: not-allowed;
-        opacity: 30%;
+      position: relative;
+      display: block;
+      width: fit-content;
+      border-radius: var(--radius-field);
+      background-color: var(--color-base-300);
+      padding-inline: calc(0.25rem * 4);
+      padding-block: calc(0.25rem * 2);
+      color: var(--color-base-content);
+      grid-row-end: 3;
+      min-height: 2rem;
+      min-width: 2.5rem;
+      max-width: 90%;
+      &:before {
+        position: absolute;
+        bottom: calc(0.25rem * 0);
+        height: calc(0.25rem * 3);
+        width: calc(0.25rem * 3);
+        background-color: inherit;
+        content: "";
+        mask-repeat: no-repeat;
+        mask-image: var(--mask-chat);
+        mask-position: 0px -1px;
+        mask-size: 0.8125rem;
      }
    }
  }
@@ -1539,6 +1525,81 @@
      padding: calc(0.25rem * 4);
    }
  }
+  .textarea {
+    @layer daisyui.l1.l2.l3 {
+      border: var(--border) solid #0000;
+      min-height: calc(0.25rem * 20);
+      flex-shrink: 1;
+      appearance: none;
+      border-radius: var(--radius-field);
+      background-color: var(--color-base-100);
+      padding-block: calc(0.25rem * 2);
+      vertical-align: middle;
+      width: clamp(3rem, 20rem, 100%);
+      padding-inline-start: 0.75rem;
+      padding-inline-end: 0.75rem;
+      font-size: max(var(--font-size, 0.875rem), 0.875rem);
+      touch-action: manipulation;
+      border-color: var(--input-color);
+      box-shadow: 0 1px var(--input-color) inset, 0 -1px oklch(100% 0 0 / calc(var(--depth) * 0.1)) inset;
+      @supports (color: color-mix(in lab, red, red)) {
+        box-shadow: 0 1px color-mix(in oklab, var(--input-color) calc(var(--depth) * 10%), #0000) inset, 0 -1px oklch(100% 0 0 / calc(var(--depth) * 0.1)) inset;
+      }
+      --input-color: var(--color-base-content);
+      @supports (color: color-mix(in lab, red, red)) {
+        --input-color: color-mix(in oklab, var(--color-base-content) 20%, #0000);
+      }
+      textarea {
+        appearance: none;
+        background-color: transparent;
+        border: none;
+        &:focus, &:focus-within {
+          --tw-outline-style: none;
+          outline-style: none;
+          @media (forced-colors: active) {
+            outline: 2px solid transparent;
+            outline-offset: 2px;
+          }
+        }
+      }
+      &:focus, &:focus-within {
+        --input-color: var(--color-base-content);
+        box-shadow: 0 1px var(--input-color);
+        @supports (color: color-mix(in lab, red, red)) {
+          box-shadow: 0 1px color-mix(in oklab, var(--input-color) calc(var(--depth) * 10%), #0000);
+        }
+        outline: 2px solid var(--input-color);
+        outline-offset: 2px;
+        isolation: isolate;
+      }
+      @media (pointer: coarse) {
+        @supports (-webkit-touch-callout: none) {
+          &:focus, &:focus-within {
+            --font-size: 1rem;
+          }
+        }
+      }
+      &:has(> textarea[disabled]), &:is(:disabled, [disabled]) {
+        cursor: not-allowed;
+        border-color: var(--color-base-200);
+        background-color: var(--color-base-200);
+        color: var(--color-base-content);
+        @supports (color: color-mix(in lab, red, red)) {
+          color: color-mix(in oklab, var(--color-base-content) 40%, transparent);
+        }
+        &::placeholder {
+          color: var(--color-base-content);
+          @supports (color: color-mix(in lab, red, red)) {
+            color: color-mix(in oklab, var(--color-base-content) 20%, transparent);
+          }
+        }
+        box-shadow: none;
+      }
+      &:has(> textarea[disabled]) > textarea[disabled] {
+        cursor: not-allowed;
+      }
+    }
+  }
  .stack {
    @layer daisyui.l1.l2.l3 {
      display: inline-grid;
--- a/docker-compose.yml
+++ b/docker-compose.yml
@@ -55,8 +55,6 @@ services:
      mongo:
        condition: service_started
    environment:
-      # LiteLLM API key (used by librechat.yaml endpoint config)
-      LITELLM_API_KEY: ${LITELLM_API_KEY:-}
      # MongoDB (use localhost since we're on host network)
      MONGO_URI: mongodb://root:example@localhost:27017/librechat?authSource=admin
      DOMAIN_CLIENT: http://localhost:3080
@@ -72,6 +70,7 @@ services:
      OPENID_CALLBACK_URL: /oauth/openid/callback
      OPENID_SCOPE: openid profile email
      OPENID_BUTTON_LABEL: Login with CERTifAI
+      OPENID_AUTH_EXTRA_PARAMS: prompt=none
      # Disable local auth (SSO only)
      ALLOW_EMAIL_LOGIN: "false"
      ALLOW_REGISTRATION: "false"
--- a/librechat/librechat.yaml
+++ b/librechat/librechat.yaml
@@ -1,5 +1,5 @@
 # CERTifAI LibreChat Configuration
-# LiteLLM proxy for unified multi-provider LLM access.
+# Ollama backend for self-hosted LLM inference.
 version: 1.2.8

 cache: true
@@ -19,16 +19,22 @@ interface:

 endpoints:
  custom:
-    - name: "LiteLLM"
-      apiKey: "${LITELLM_API_KEY}"
-      baseURL: "https://llm-dev.meghsakha.com/v1/"
+    - name: "Ollama"
+      apiKey: "ollama"
+      baseURL: "https://mac-mini-von-benjamin-2:11434/v1/"
      models:
        default:
-          - "Qwen3-Coder-30B-A3B-Instruct"
+          - "llama3.1:8b"
+          - "qwen3:30b-a3b"
        fetch: true
      titleConvo: true
      titleModel: "current_model"
      summarize: false
      summaryModel: "current_model"
      forcePrompt: false
-      modelDisplayLabel: "CERTifAI LiteLLM"
+      modelDisplayLabel: "CERTifAI Ollama"
+      dropParams:
+        - stop
+        - user
+        - frequency_penalty
+        - presence_penalty
--- a/src/components/dashboard_sidebar.rs
+++ b/src/components/dashboard_sidebar.rs
@@ -1,9 +1,9 @@
 use dioxus::prelude::*;

 use crate::i18n::{t, Locale};
-use crate::infrastructure::litellm::{get_litellm_status, LitellmStatus};
+use crate::infrastructure::ollama::{get_ollama_status, OllamaStatus};

-/// Right sidebar for the dashboard, showing LiteLLM status, trending topics,
+/// Right sidebar for the dashboard, showing Ollama status, trending topics,
 /// and recent search history.
 ///
 /// Appears when no article card is selected. Disappears when the user opens
@@ -11,13 +11,13 @@ use crate::infrastructure::litellm::{get_litellm_status, LitellmStatus};
 ///
 /// # Props
 ///
-/// * `litellm_url` - LiteLLM proxy URL for status polling
+/// * `ollama_url` - Ollama instance URL for status polling
 /// * `trending` - Trending topic keywords extracted from recent news headlines
 /// * `recent_searches` - Recent search topics stored in localStorage
 /// * `on_topic_click` - Fires when a trending or recent topic is clicked
 #[component]
 pub fn DashboardSidebar(
-    litellm_url: String,
+    ollama_url: String,
    trending: Vec<String>,
    recent_searches: Vec<String>,
    on_topic_click: EventHandler<String>,
@@ -25,26 +25,26 @@ pub fn DashboardSidebar(
    let locale = use_context::<Signal<Locale>>();
    let l = *locale.read();

-    // Fetch LiteLLM status once on mount.
+    // Fetch Ollama status once on mount.
    // use_resource with no signal dependencies runs exactly once and
    // won't re-fire on parent re-renders (unlike use_effect).
-    let url = litellm_url.clone();
+    let url = ollama_url.clone();
    let status_resource = use_resource(move || {
        let u = url.clone();
        async move {
-            get_litellm_status(u).await.unwrap_or(LitellmStatus {
+            get_ollama_status(u).await.unwrap_or(OllamaStatus {
                online: false,
                models: Vec::new(),
            })
        }
    });

-    let current_status: LitellmStatus =
+    let current_status: OllamaStatus =
        status_resource
            .read()
            .as_ref()
            .cloned()
-            .unwrap_or(LitellmStatus {
+            .unwrap_or(OllamaStatus {
                online: false,
                models: Vec::new(),
            });
@@ -52,9 +52,9 @@ pub fn DashboardSidebar(
    rsx! {
        aside { class: "dashboard-sidebar",

-            // -- LiteLLM Status Section --
+            // -- Ollama Status Section --
            div { class: "sidebar-section",
-                h4 { class: "sidebar-section-title", "{t(l, \"dashboard.litellm_status\")}" }
+                h4 { class: "sidebar-section-title", "{t(l, \"dashboard.ollama_status\")}" }
                div { class: "sidebar-status-row",
                    span { class: if current_status.online { "sidebar-status-dot sidebar-status-dot--online" } else { "sidebar-status-dot sidebar-status-dot--offline" } }
                    span { class: "sidebar-status-label",
--- a/src/components/news_card.rs
+++ b/src/components/news_card.rs
@@ -112,12 +112,12 @@ pub fn mock_news() -> Vec<NewsCardModel> {
            published_at: "2026-02-16".into(),
        },
        NewsCardModel {
-            title: "LiteLLM Adds Multi-Provider Routing".into(),
-            source: "LiteLLM".into(),
-            summary: "Route requests across multiple LLM providers with automatic fallback.".into(),
-            content: "LiteLLM now supports multi-provider routing with automatic \
-                fallback. Users can route requests across multiple providers \
-                for improved reliability and cost optimization."
+            title: "Ollama Adds Multi-GPU Scheduling".into(),
+            source: "Ollama".into(),
+            summary: "Run large models across multiple GPUs with automatic sharding.".into(),
+            content: "Ollama now supports multi-GPU scheduling with automatic \
+                model sharding. Users can run models across multiple GPUs \
+                for improved inference performance."
                .into(),
            category: "Infrastructure".into(),
            url: "#".into(),
--- a/src/infrastructure/chat.rs
+++ b/src/infrastructure/chat.rs
@@ -134,7 +134,7 @@ pub async fn list_chat_sessions() -> Result<Vec<ChatSession>, ServerFnError> {
 ///
 /// * `title` - Display title for the session
 /// * `namespace` - Namespace string: `"General"` or `"News"`
-/// * `provider` - LLM provider name (e.g. "litellm")
+/// * `provider` - LLM provider name (e.g. "ollama")
 /// * `model` - Model ID (e.g. "llama3.1:8b")
 /// * `article_url` - Source article URL (only for `News` namespace, empty if none)
 ///
@@ -441,8 +441,8 @@ pub async fn chat_complete(

    // Resolve provider URL and model
    let (base_url, model) = resolve_provider_url(
-        &state.services.litellm_url,
-        &state.services.litellm_model,
+        &state.services.ollama_url,
+        &state.services.ollama_model,
        &session.provider,
        &session.model,
    );
@@ -485,22 +485,22 @@ pub async fn chat_complete(
        .ok_or_else(|| ServerFnError::new("empty LLM response"))
 }

-/// Resolve the base URL for a provider, falling back to LiteLLM defaults.
+/// Resolve the base URL for a provider, falling back to Ollama defaults.
 ///
 /// # Arguments
 ///
-/// * `litellm_url` - Default LiteLLM base URL from config
-/// * `litellm_model` - Default LiteLLM model from config
+/// * `ollama_url` - Default Ollama base URL from config
+/// * `ollama_model` - Default Ollama model from config
 /// * `provider` - Provider name (e.g. "openai", "anthropic", "huggingface")
-/// * `model` - Model ID (may be empty for LiteLLM default)
+/// * `model` - Model ID (may be empty for Ollama default)
 ///
 /// # Returns
 ///
 /// A `(base_url, model)` tuple resolved for the given provider.
 #[cfg(feature = "server")]
 pub(crate) fn resolve_provider_url(
-    litellm_url: &str,
-    litellm_model: &str,
+    ollama_url: &str,
+    ollama_model: &str,
    provider: &str,
    model: &str,
 ) -> (String, String) {
@@ -511,11 +511,11 @@ pub(crate) fn resolve_provider_url(
            format!("https://api-inference.huggingface.co/models/{}", model),
            model.to_string(),
        ),
-        // Default to LiteLLM
+        // Default to Ollama
        _ => (
-            litellm_url.to_string(),
+            ollama_url.to_string(),
            if model.is_empty() {
-                litellm_model.to_string()
+                ollama_model.to_string()
            } else {
                model.to_string()
            },
@@ -595,7 +595,7 @@ mod tests {
                "_id": oid,
                "user_sub": "u",
                "title": "t",
-                "provider": "litellm",
+                "provider": "ollama",
                "model": "m",
                "created_at": "c",
                "updated_at": "u",
@@ -612,7 +612,7 @@ mod tests {
                "user_sub": "u",
                "title": "t",
                "namespace": "News",
-                "provider": "litellm",
+                "provider": "ollama",
                "model": "m",
                "created_at": "c",
                "updated_at": "u",
@@ -684,13 +684,13 @@ mod tests {

        // -- resolve_provider_url --

-        const TEST_LITELLM_URL: &str = "http://localhost:4000";
-        const TEST_LITELLM_MODEL: &str = "qwen3-32b";
+        const TEST_OLLAMA_URL: &str = "http://localhost:11434";
+        const TEST_OLLAMA_MODEL: &str = "llama3.1:8b";

        #[test]
        fn resolve_openai_returns_api_openai() {
            let (url, model) =
-                resolve_provider_url(TEST_LITELLM_URL, TEST_LITELLM_MODEL, "openai", "gpt-4o");
+                resolve_provider_url(TEST_OLLAMA_URL, TEST_OLLAMA_MODEL, "openai", "gpt-4o");
            assert_eq!(url, "https://api.openai.com");
            assert_eq!(model, "gpt-4o");
        }
@@ -698,8 +698,8 @@ mod tests {
        #[test]
        fn resolve_anthropic_returns_api_anthropic() {
            let (url, model) = resolve_provider_url(
-                TEST_LITELLM_URL,
-                TEST_LITELLM_MODEL,
+                TEST_OLLAMA_URL,
+                TEST_OLLAMA_MODEL,
                "anthropic",
                "claude-3-opus",
            );
@@ -710,8 +710,8 @@ mod tests {
        #[test]
        fn resolve_huggingface_returns_model_url() {
            let (url, model) = resolve_provider_url(
-                TEST_LITELLM_URL,
-                TEST_LITELLM_MODEL,
+                TEST_OLLAMA_URL,
+                TEST_OLLAMA_MODEL,
                "huggingface",
                "meta-llama/Llama-2-7b",
            );
@@ -723,19 +723,19 @@ mod tests {
        }

        #[test]
-        fn resolve_unknown_defaults_to_litellm() {
+        fn resolve_unknown_defaults_to_ollama() {
            let (url, model) =
-                resolve_provider_url(TEST_LITELLM_URL, TEST_LITELLM_MODEL, "litellm", "qwen3-32b");
-            assert_eq!(url, TEST_LITELLM_URL);
-            assert_eq!(model, "qwen3-32b");
+                resolve_provider_url(TEST_OLLAMA_URL, TEST_OLLAMA_MODEL, "ollama", "mistral:7b");
+            assert_eq!(url, TEST_OLLAMA_URL);
+            assert_eq!(model, "mistral:7b");
        }

        #[test]
        fn resolve_empty_model_falls_back_to_server_default() {
            let (url, model) =
-                resolve_provider_url(TEST_LITELLM_URL, TEST_LITELLM_MODEL, "litellm", "");
-            assert_eq!(url, TEST_LITELLM_URL);
-            assert_eq!(model, TEST_LITELLM_MODEL);
+                resolve_provider_url(TEST_OLLAMA_URL, TEST_OLLAMA_MODEL, "ollama", "");
+            assert_eq!(url, TEST_OLLAMA_URL);
+            assert_eq!(model, TEST_OLLAMA_MODEL);
        }
    }
 }
--- a/src/infrastructure/config.rs
+++ b/src/infrastructure/config.rs
@@ -141,15 +141,13 @@ impl SmtpConfig {
 // ServiceUrls
 // ---------------------------------------------------------------------------

-/// URLs and credentials for external services (LiteLLM, SearXNG, S3, etc.).
+/// URLs and credentials for external services (Ollama, SearXNG, S3, etc.).
 #[derive(Debug)]
 pub struct ServiceUrls {
-    /// LiteLLM proxy base URL.
-    pub litellm_url: String,
-    /// Default LiteLLM model to use.
-    pub litellm_model: String,
-    /// LiteLLM API key for authenticated requests.
-    pub litellm_api_key: String,
+    /// Ollama LLM instance base URL.
+    pub ollama_url: String,
+    /// Default Ollama model to use.
+    pub ollama_model: String,
    /// SearXNG meta-search engine base URL.
    pub searxng_url: String,
    /// LangChain service URL.
@@ -180,10 +178,9 @@ impl ServiceUrls {
    /// Currently infallible but returns `Result` for consistency.
    pub fn from_env() -> Result<Self, Error> {
        Ok(Self {
-            litellm_url: std::env::var("LITELLM_URL")
-                .unwrap_or_else(|_| "http://localhost:4000".into()),
-            litellm_model: std::env::var("LITELLM_MODEL").unwrap_or_else(|_| "qwen3-32b".into()),
-            litellm_api_key: optional_env("LITELLM_API_KEY"),
+            ollama_url: std::env::var("OLLAMA_URL")
+                .unwrap_or_else(|_| "http://localhost:11434".into()),
+            ollama_model: std::env::var("OLLAMA_MODEL").unwrap_or_else(|_| "llama3.1:8b".into()),
            searxng_url: std::env::var("SEARXNG_URL")
                .unwrap_or_else(|_| "http://localhost:8888".into()),
            langchain_url: optional_env("LANGCHAIN_URL"),
@@ -234,7 +231,7 @@ impl StripeConfig {

 /// Comma-separated list of enabled LLM provider identifiers.
 ///
-/// For example: `LLM_PROVIDERS=litellm,openai,anthropic`
+/// For example: `LLM_PROVIDERS=ollama,openai,anthropic`
 #[derive(Debug)]
 pub struct LlmProvidersConfig {
    /// Parsed provider names.
@@ -334,36 +331,36 @@ mod tests {
    #[test]
    #[serial]
    fn llm_providers_single() {
-        std::env::set_var("LLM_PROVIDERS", "litellm");
+        std::env::set_var("LLM_PROVIDERS", "ollama");
        let cfg = LlmProvidersConfig::from_env().unwrap();
-        assert_eq!(cfg.providers, vec!["litellm"]);
+        assert_eq!(cfg.providers, vec!["ollama"]);
        std::env::remove_var("LLM_PROVIDERS");
    }

    #[test]
    #[serial]
    fn llm_providers_multiple() {
-        std::env::set_var("LLM_PROVIDERS", "litellm,openai,anthropic");
+        std::env::set_var("LLM_PROVIDERS", "ollama,openai,anthropic");
        let cfg = LlmProvidersConfig::from_env().unwrap();
-        assert_eq!(cfg.providers, vec!["litellm", "openai", "anthropic"]);
+        assert_eq!(cfg.providers, vec!["ollama", "openai", "anthropic"]);
        std::env::remove_var("LLM_PROVIDERS");
    }

    #[test]
    #[serial]
    fn llm_providers_trims_whitespace() {
-        std::env::set_var("LLM_PROVIDERS", " litellm , openai ");
+        std::env::set_var("LLM_PROVIDERS", " ollama , openai ");
        let cfg = LlmProvidersConfig::from_env().unwrap();
-        assert_eq!(cfg.providers, vec!["litellm", "openai"]);
+        assert_eq!(cfg.providers, vec!["ollama", "openai"]);
        std::env::remove_var("LLM_PROVIDERS");
    }

    #[test]
    #[serial]
    fn llm_providers_filters_empty_entries() {
-        std::env::set_var("LLM_PROVIDERS", "litellm,,openai,");
+        std::env::set_var("LLM_PROVIDERS", "ollama,,openai,");
        let cfg = LlmProvidersConfig::from_env().unwrap();
-        assert_eq!(cfg.providers, vec!["litellm", "openai"]);
+        assert_eq!(cfg.providers, vec!["ollama", "openai"]);
        std::env::remove_var("LLM_PROVIDERS");
    }

@@ -373,18 +370,18 @@ mod tests {

    #[test]
    #[serial]
-    fn service_urls_default_litellm_url() {
-        std::env::remove_var("LITELLM_URL");
+    fn service_urls_default_ollama_url() {
+        std::env::remove_var("OLLAMA_URL");
        let svc = ServiceUrls::from_env().unwrap();
-        assert_eq!(svc.litellm_url, "http://localhost:4000");
+        assert_eq!(svc.ollama_url, "http://localhost:11434");
    }

    #[test]
    #[serial]
-    fn service_urls_default_litellm_model() {
-        std::env::remove_var("LITELLM_MODEL");
+    fn service_urls_default_ollama_model() {
+        std::env::remove_var("OLLAMA_MODEL");
        let svc = ServiceUrls::from_env().unwrap();
-        assert_eq!(svc.litellm_model, "qwen3-32b");
+        assert_eq!(svc.ollama_model, "llama3.1:8b");
    }

    #[test]
@@ -397,11 +394,11 @@ mod tests {

    #[test]
    #[serial]
-    fn service_urls_custom_litellm_url() {
-        std::env::set_var("LITELLM_URL", "http://litellm-host:4000");
+    fn service_urls_custom_ollama_url() {
+        std::env::set_var("OLLAMA_URL", "http://gpu-host:11434");
        let svc = ServiceUrls::from_env().unwrap();
-        assert_eq!(svc.litellm_url, "http://litellm-host:4000");
-        std::env::remove_var("LITELLM_URL");
+        assert_eq!(svc.ollama_url, "http://gpu-host:11434");
+        std::env::remove_var("OLLAMA_URL");
    }

    #[test]
--- a/src/infrastructure/litellm.rs
+++ b/src/infrastructure/litellm.rs
@@ -1,403 +0,0 @@
-#[cfg(feature = "server")]
-use std::collections::HashMap;
-
-use dioxus::prelude::*;
-use serde::{Deserialize, Serialize};
-
-use crate::models::LitellmUsageStats;
-#[cfg(feature = "server")]
-use crate::models::ModelUsage;
-
-/// Status of a LiteLLM proxy instance, including connectivity and available models.
-///
-/// # Fields
-///
-/// * `online` - Whether the LiteLLM API responded successfully
-/// * `models` - List of model IDs available through the proxy
-#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)]
-pub struct LitellmStatus {
-    pub online: bool,
-    pub models: Vec<String>,
-}
-
-/// Response from LiteLLM's `GET /v1/models` endpoint (OpenAI-compatible).
-#[cfg(feature = "server")]
-#[derive(Deserialize)]
-struct ModelsResponse {
-    data: Vec<ModelObject>,
-}
-
-/// A single model entry from the OpenAI-compatible models list.
-#[cfg(feature = "server")]
-#[derive(Deserialize)]
-struct ModelObject {
-    id: String,
-}
-
-/// Check the status of a LiteLLM proxy by querying its models endpoint.
-///
-/// Calls `GET <litellm_url>/v1/models` to list available models and determine
-/// whether the instance is reachable. Sends the API key as a Bearer token
-/// if configured.
-///
-/// # Arguments
-///
-/// * `litellm_url` - Base URL of the LiteLLM proxy (e.g. "http://localhost:4000")
-///
-/// # Returns
-///
-/// A `LitellmStatus` with `online: true` and model IDs if reachable,
-/// or `online: false` with an empty model list on failure
-///
-/// # Errors
-///
-/// Returns `ServerFnError` only on serialization issues; network failures
-/// are caught and returned as `online: false`
-#[post("/api/litellm-status")]
-pub async fn get_litellm_status(litellm_url: String) -> Result<LitellmStatus, ServerFnError> {
-    let state: crate::infrastructure::ServerState =
-        dioxus_fullstack::FullstackContext::extract().await?;
-
-    let base_url = if litellm_url.is_empty() {
-        state.services.litellm_url.clone()
-    } else {
-        litellm_url
-    };
-
-    let api_key = state.services.litellm_api_key.clone();
-    let url = format!("{}/v1/models", base_url.trim_end_matches('/'));
-
-    let client = reqwest::Client::builder()
-        .timeout(std::time::Duration::from_secs(5))
-        .build()
-        .map_err(|e| ServerFnError::new(format!("HTTP client error: {e}")))?;
-
-    let mut request = client.get(&url);
-    if !api_key.is_empty() {
-        request = request.header("Authorization", format!("Bearer {api_key}"));
-    }
-
-    let resp = match request.send().await {
-        Ok(r) if r.status().is_success() => r,
-        _ => {
-            return Ok(LitellmStatus {
-                online: false,
-                models: Vec::new(),
-            });
-        }
-    };
-
-    let body: ModelsResponse = match resp.json().await {
-        Ok(b) => b,
-        Err(_) => {
-            return Ok(LitellmStatus {
-                online: true,
-                models: Vec::new(),
-            });
-        }
-    };
-
-    let models = body.data.into_iter().map(|m| m.id).collect();
-
-    Ok(LitellmStatus {
-        online: true,
-        models,
-    })
-}
-
-/// Response from LiteLLM's `GET /global/activity` endpoint.
-///
-/// Returns aggregate token counts and API request totals for a date range.
-/// Available on the free tier (no Enterprise license needed).
-#[cfg(feature = "server")]
-#[derive(Debug, Deserialize)]
-struct ActivityResponse {
-    /// Total tokens across all models in the date range
-    #[serde(default)]
-    sum_total_tokens: u64,
-}
-
-/// Per-model entry from `GET /global/activity/model`.
-///
-/// Each entry contains a model name and its aggregated token total.
-#[cfg(feature = "server")]
-#[derive(Debug, Deserialize)]
-struct ActivityModelEntry {
-    /// Model identifier (may be empty for unattributed traffic)
-    #[serde(default)]
-    model: String,
-    /// Sum of tokens used by this model in the date range
-    #[serde(default)]
-    sum_total_tokens: u64,
-}
-
-/// Per-model spend entry from `GET /global/spend/models`.
-///
-/// Each entry maps a model name to its total spend in USD.
-#[cfg(feature = "server")]
-#[derive(Debug, Deserialize)]
-struct SpendModelEntry {
-    /// Model identifier
-    #[serde(default)]
-    model: String,
-    /// Total spend in USD
-    #[serde(default)]
-    total_spend: f64,
-}
-
-/// Merge per-model token counts and spend data into `ModelUsage` entries.
-///
-/// Joins `activity_models` (tokens) and `spend_models` (spend) by model
-/// name using a HashMap for O(n + m) merge. Entries with empty model
-/// names are skipped.
-///
-/// # Arguments
-///
-/// * `activity_models` - Per-model token data from `/global/activity/model`
-/// * `spend_models` - Per-model spend data from `/global/spend/models`
-///
-/// # Returns
-///
-/// Merged list sorted by total tokens descending
-#[cfg(feature = "server")]
-fn merge_model_data(
-    activity_models: Vec<ActivityModelEntry>,
-    spend_models: Vec<SpendModelEntry>,
-) -> Vec<ModelUsage> {
-    let mut model_map: HashMap<String, ModelUsage> = HashMap::new();
-
-    for entry in activity_models {
-        if entry.model.is_empty() {
-            continue;
-        }
-        model_map
-            .entry(entry.model.clone())
-            .or_insert_with(|| ModelUsage {
-                model: entry.model,
-                ..Default::default()
-            })
-            .total_tokens = entry.sum_total_tokens;
-    }
-
-    for entry in spend_models {
-        if entry.model.is_empty() {
-            continue;
-        }
-        model_map
-            .entry(entry.model.clone())
-            .or_insert_with(|| ModelUsage {
-                model: entry.model,
-                ..Default::default()
-            })
-            .spend = entry.total_spend;
-    }
-
-    let mut result: Vec<ModelUsage> = model_map.into_values().collect();
-    result.sort_by(|a, b| b.total_tokens.cmp(&a.total_tokens));
-    result
-}
-
-/// Fetch aggregated usage statistics from LiteLLM's free-tier APIs.
-///
-/// Combines three endpoints to build a complete usage picture:
-/// - `GET /global/activity` - total token counts
-/// - `GET /global/activity/model` - per-model token breakdown
-/// - `GET /global/spend/models` - per-model spend in USD
-///
-/// # Arguments
-///
-/// * `start_date` - Start of the reporting period in `YYYY-MM-DD` format
-/// * `end_date` - End of the reporting period in `YYYY-MM-DD` format
-///
-/// # Returns
-///
-/// Aggregated usage stats; returns default (zeroed) stats on network
-/// failure or permission errors
-///
-/// # Errors
-///
-/// Returns `ServerFnError` only on HTTP client construction failure
-#[post("/api/litellm-usage")]
-pub async fn get_litellm_usage(
-    start_date: String,
-    end_date: String,
-) -> Result<LitellmUsageStats, ServerFnError> {
-    let state: crate::infrastructure::ServerState =
-        dioxus_fullstack::FullstackContext::extract().await?;
-
-    let base_url = &state.services.litellm_url;
-    let api_key = &state.services.litellm_api_key;
-
-    if base_url.is_empty() {
-        return Ok(LitellmUsageStats::default());
-    }
-
-    let base = base_url.trim_end_matches('/');
-    let date_params = format!("start_date={start_date}&end_date={end_date}");
-
-    let client = reqwest::Client::builder()
-        .timeout(std::time::Duration::from_secs(10))
-        .build()
-        .map_err(|e| ServerFnError::new(format!("HTTP client error: {e}")))?;
-
-    // Helper closure to build an authenticated GET request
-    let auth_get = |url: String| {
-        let mut req = client.get(url);
-        if !api_key.is_empty() {
-            req = req.header("Authorization", format!("Bearer {api_key}"));
-        }
-        req
-    };
-
-    // Fire all three requests concurrently to minimise latency
-    let (activity_res, model_activity_res, model_spend_res) = tokio::join!(
-        auth_get(format!("{base}/global/activity?{date_params}")).send(),
-        auth_get(format!("{base}/global/activity/model?{date_params}")).send(),
-        auth_get(format!("{base}/global/spend/models?{date_params}")).send(),
-    );
-
-    // Parse total token count from /global/activity
-    let total_tokens = match activity_res {
-        Ok(r) if r.status().is_success() => r
-            .json::<ActivityResponse>()
-            .await
-            .map(|a| a.sum_total_tokens)
-            .unwrap_or(0),
-        _ => 0,
-    };
-
-    // Parse per-model token breakdown from /global/activity/model
-    let activity_models: Vec<ActivityModelEntry> = match model_activity_res {
-        Ok(r) if r.status().is_success() => r.json().await.unwrap_or_default(),
-        _ => Vec::new(),
-    };
-
-    // Parse per-model spend from /global/spend/models
-    let spend_models: Vec<SpendModelEntry> = match model_spend_res {
-        Ok(r) if r.status().is_success() => r.json().await.unwrap_or_default(),
-        _ => Vec::new(),
-    };
-
-    let total_spend: f64 = spend_models.iter().map(|m| m.total_spend).sum();
-    let model_breakdown = merge_model_data(activity_models, spend_models);
-
-    Ok(LitellmUsageStats {
-        total_spend,
-        // Free-tier endpoints don't provide prompt/completion split;
-        // total_tokens comes from /global/activity.
-        total_prompt_tokens: 0,
-        total_completion_tokens: 0,
-        total_tokens,
-        model_breakdown,
-    })
-}
-
-#[cfg(all(test, feature = "server"))]
-mod tests {
-    use super::*;
-
-    #[test]
-    fn merge_empty_inputs() {
-        let result = merge_model_data(Vec::new(), Vec::new());
-        assert!(result.is_empty());
-    }
-
-    #[test]
-    fn merge_activity_only() {
-        let activity = vec![ActivityModelEntry {
-            model: "gpt-4".into(),
-            sum_total_tokens: 1500,
-        }];
-        let result = merge_model_data(activity, Vec::new());
-        assert_eq!(result.len(), 1);
-        assert_eq!(result[0].model, "gpt-4");
-        assert_eq!(result[0].total_tokens, 1500);
-        assert_eq!(result[0].spend, 0.0);
-    }
-
-    #[test]
-    fn merge_spend_only() {
-        let spend = vec![SpendModelEntry {
-            model: "gpt-4".into(),
-            total_spend: 2.5,
-        }];
-        let result = merge_model_data(Vec::new(), spend);
-        assert_eq!(result.len(), 1);
-        assert_eq!(result[0].model, "gpt-4");
-        assert_eq!(result[0].spend, 2.5);
-        assert_eq!(result[0].total_tokens, 0);
-    }
-
-    #[test]
-    fn merge_joins_by_model_name() {
-        let activity = vec![
-            ActivityModelEntry {
-                model: "gpt-4".into(),
-                sum_total_tokens: 5000,
-            },
-            ActivityModelEntry {
-                model: "claude-3".into(),
-                sum_total_tokens: 3000,
-            },
-        ];
-        let spend = vec![
-            SpendModelEntry {
-                model: "gpt-4".into(),
-                total_spend: 1.0,
-            },
-            SpendModelEntry {
-                model: "claude-3".into(),
-                total_spend: 0.5,
-            },
-        ];
-        let result = merge_model_data(activity, spend);
-        assert_eq!(result.len(), 2);
-        // Sorted by tokens descending: gpt-4 (5000) before claude-3 (3000)
-        assert_eq!(result[0].model, "gpt-4");
-        assert_eq!(result[0].total_tokens, 5000);
-        assert_eq!(result[0].spend, 1.0);
-        assert_eq!(result[1].model, "claude-3");
-        assert_eq!(result[1].total_tokens, 3000);
-        assert_eq!(result[1].spend, 0.5);
-    }
-
-    #[test]
-    fn merge_skips_empty_model_names() {
-        let activity = vec![
-            ActivityModelEntry {
-                model: "".into(),
-                sum_total_tokens: 100,
-            },
-            ActivityModelEntry {
-                model: "gpt-4".into(),
-                sum_total_tokens: 500,
-            },
-        ];
-        let spend = vec![SpendModelEntry {
-            model: "".into(),
-            total_spend: 0.01,
-        }];
-        let result = merge_model_data(activity, spend);
-        assert_eq!(result.len(), 1);
-        assert_eq!(result[0].model, "gpt-4");
-    }
-
-    #[test]
-    fn merge_unmatched_models_appear_in_both_directions() {
-        let activity = vec![ActivityModelEntry {
-            model: "tokens-only".into(),
-            sum_total_tokens: 1000,
-        }];
-        let spend = vec![SpendModelEntry {
-            model: "spend-only".into(),
-            total_spend: 0.5,
-        }];
-        let result = merge_model_data(activity, spend);
-        assert_eq!(result.len(), 2);
-        // tokens-only has 1000 tokens, spend-only has 0 tokens
-        assert_eq!(result[0].model, "tokens-only");
-        assert_eq!(result[0].total_tokens, 1000);
-        assert_eq!(result[1].model, "spend-only");
-        assert_eq!(result[1].spend, 0.5);
-    }
-}
--- a/src/infrastructure/llm.rs
+++ b/src/infrastructure/llm.rs
@@ -4,23 +4,23 @@ use dioxus::prelude::*;
 mod inner {
    use serde::{Deserialize, Serialize};

-    /// A single message in the OpenAI-compatible chat format used by LiteLLM.
+    /// A single message in the OpenAI-compatible chat format used by Ollama.
    #[derive(Serialize)]
    pub(super) struct ChatMessage {
        pub role: String,
        pub content: String,
    }

-    /// Request body for the OpenAI-compatible chat completions endpoint.
+    /// Request body for Ollama's OpenAI-compatible chat completions endpoint.
    #[derive(Serialize)]
-    pub(super) struct ChatCompletionRequest {
+    pub(super) struct OllamaChatRequest {
        pub model: String,
        pub messages: Vec<ChatMessage>,
        /// Disable streaming so we get a single JSON response.
        pub stream: bool,
    }

-    /// A single choice in the chat completions response.
+    /// A single choice in the Ollama chat completions response.
    #[derive(Deserialize)]
    pub(super) struct ChatChoice {
        pub message: ChatResponseMessage,
@@ -32,9 +32,9 @@ mod inner {
        pub content: String,
    }

-    /// Top-level response from the `/v1/chat/completions` endpoint.
+    /// Top-level response from Ollama's `/v1/chat/completions` endpoint.
    #[derive(Deserialize)]
-    pub(super) struct ChatCompletionResponse {
+    pub(super) struct OllamaChatResponse {
        pub choices: Vec<ChatChoice>,
    }

@@ -157,7 +157,7 @@ mod inner {
    }
 }

-/// Summarize an article using a LiteLLM proxy.
+/// Summarize an article using a local Ollama instance.
 ///
 /// First attempts to fetch the full article text from the provided URL.
 /// If that fails (paywall, timeout, etc.), falls back to the search snippet.
@@ -167,8 +167,8 @@ mod inner {
 ///
 /// * `snippet` - The search result snippet (fallback content)
 /// * `article_url` - The original article URL to fetch full text from
-/// * `litellm_url` - Base URL of the LiteLLM proxy (e.g. "http://localhost:4000")
-/// * `model` - The model ID to use (e.g. "qwen3-32b")
+/// * `ollama_url` - Base URL of the Ollama instance (e.g. "http://localhost:11434")
+/// * `model` - The Ollama model ID to use (e.g. "llama3.1:8b")
 ///
 /// # Returns
 ///
@@ -176,38 +176,36 @@ mod inner {
 ///
 /// # Errors
 ///
-/// Returns `ServerFnError` if the LiteLLM request fails or response parsing fails
+/// Returns `ServerFnError` if the Ollama request fails or response parsing fails
 #[post("/api/summarize")]
 pub async fn summarize_article(
    snippet: String,
    article_url: String,
-    litellm_url: String,
+    ollama_url: String,
    model: String,
 ) -> Result<String, ServerFnError> {
-    use inner::{fetch_article_text, ChatCompletionRequest, ChatCompletionResponse, ChatMessage};
+    use inner::{fetch_article_text, ChatMessage, OllamaChatRequest, OllamaChatResponse};

    let state: crate::infrastructure::ServerState =
        dioxus_fullstack::FullstackContext::extract().await?;

    // Use caller-provided values or fall back to ServerState config
-    let base_url = if litellm_url.is_empty() {
-        state.services.litellm_url.clone()
+    let base_url = if ollama_url.is_empty() {
+        state.services.ollama_url.clone()
    } else {
-        litellm_url
+        ollama_url
    };

    let model = if model.is_empty() {
-        state.services.litellm_model.clone()
+        state.services.ollama_model.clone()
    } else {
        model
    };

-    let api_key = state.services.litellm_api_key.clone();
-
    // Try to fetch the full article; fall back to the search snippet
    let article_text = fetch_article_text(&article_url).await.unwrap_or(snippet);

-    let request_body = ChatCompletionRequest {
+    let request_body = OllamaChatRequest {
        model,
        stream: false,
        messages: vec![ChatMessage {
@@ -225,48 +223,42 @@ pub async fn summarize_article(

    let url = format!("{}/v1/chat/completions", base_url.trim_end_matches('/'));
    let client = reqwest::Client::new();
-    let mut request = client
+    let resp = client
        .post(&url)
        .header("content-type", "application/json")
-        .json(&request_body);
-
-    if !api_key.is_empty() {
-        request = request.header("Authorization", format!("Bearer {api_key}"));
-    }
-
-    let resp = request
+        .json(&request_body)
        .send()
        .await
-        .map_err(|e| ServerFnError::new(format!("LiteLLM request failed: {e}")))?;
+        .map_err(|e| ServerFnError::new(format!("Ollama request failed: {e}")))?;

    if !resp.status().is_success() {
        let status = resp.status();
        let body = resp.text().await.unwrap_or_default();
        return Err(ServerFnError::new(format!(
-            "LiteLLM returned {status}: {body}"
+            "Ollama returned {status}: {body}"
        )));
    }

-    let body: ChatCompletionResponse = resp
+    let body: OllamaChatResponse = resp
        .json()
        .await
-        .map_err(|e| ServerFnError::new(format!("Failed to parse LiteLLM response: {e}")))?;
+        .map_err(|e| ServerFnError::new(format!("Failed to parse Ollama response: {e}")))?;

    body.choices
        .first()
        .map(|choice| choice.message.content.clone())
-        .ok_or_else(|| ServerFnError::new("Empty response from LiteLLM"))
+        .ok_or_else(|| ServerFnError::new("Empty response from Ollama"))
 }

 /// A lightweight chat message for the follow-up conversation.
-/// Uses simple String role ("system"/"user"/"assistant") for OpenAI compatibility.
+/// Uses simple String role ("system"/"user"/"assistant") for Ollama compatibility.
 #[derive(Debug, Clone, PartialEq, serde::Serialize, serde::Deserialize)]
 pub struct FollowUpMessage {
    pub role: String,
    pub content: String,
 }

-/// Send a follow-up question about an article using a LiteLLM proxy.
+/// Send a follow-up question about an article using a local Ollama instance.
 ///
 /// Accepts the full conversation history (system context + prior turns) and
 /// returns the assistant's next response. The system message should contain
@@ -275,8 +267,8 @@ pub struct FollowUpMessage {
 /// # Arguments
 ///
 /// * `messages` - The conversation history including system context
-/// * `litellm_url` - Base URL of the LiteLLM proxy
-/// * `model` - The model ID to use
+/// * `ollama_url` - Base URL of the Ollama instance
+/// * `model` - The Ollama model ID to use
 ///
 /// # Returns
 ///
@@ -284,32 +276,30 @@ pub struct FollowUpMessage {
 ///
 /// # Errors
 ///
-/// Returns `ServerFnError` if the LiteLLM request fails or response parsing fails
+/// Returns `ServerFnError` if the Ollama request fails or response parsing fails
 #[post("/api/chat")]
 pub async fn chat_followup(
    messages: Vec<FollowUpMessage>,
-    litellm_url: String,
+    ollama_url: String,
    model: String,
 ) -> Result<String, ServerFnError> {
-    use inner::{ChatCompletionRequest, ChatCompletionResponse, ChatMessage};
+    use inner::{ChatMessage, OllamaChatRequest, OllamaChatResponse};

    let state: crate::infrastructure::ServerState =
        dioxus_fullstack::FullstackContext::extract().await?;

-    let base_url = if litellm_url.is_empty() {
-        state.services.litellm_url.clone()
+    let base_url = if ollama_url.is_empty() {
+        state.services.ollama_url.clone()
    } else {
-        litellm_url
+        ollama_url
    };

    let model = if model.is_empty() {
-        state.services.litellm_model.clone()
+        state.services.ollama_model.clone()
    } else {
        model
    };

-    let api_key = state.services.litellm_api_key.clone();
-
    // Convert FollowUpMessage to inner ChatMessage for the request
    let chat_messages: Vec<ChatMessage> = messages
        .into_iter()
@@ -319,7 +309,7 @@ pub async fn chat_followup(
        })
        .collect();

-    let request_body = ChatCompletionRequest {
+    let request_body = OllamaChatRequest {
        model,
        stream: false,
        messages: chat_messages,
@@ -327,37 +317,31 @@ pub async fn chat_followup(

    let url = format!("{}/v1/chat/completions", base_url.trim_end_matches('/'));
    let client = reqwest::Client::new();
-    let mut request = client
+    let resp = client
        .post(&url)
        .header("content-type", "application/json")
-        .json(&request_body);
-
-    if !api_key.is_empty() {
-        request = request.header("Authorization", format!("Bearer {api_key}"));
-    }
-
-    let resp = request
+        .json(&request_body)
        .send()
        .await
-        .map_err(|e| ServerFnError::new(format!("LiteLLM request failed: {e}")))?;
+        .map_err(|e| ServerFnError::new(format!("Ollama request failed: {e}")))?;

    if !resp.status().is_success() {
        let status = resp.status();
        let body = resp.text().await.unwrap_or_default();
        return Err(ServerFnError::new(format!(
-            "LiteLLM returned {status}: {body}"
+            "Ollama returned {status}: {body}"
        )));
    }

-    let body: ChatCompletionResponse = resp
+    let body: OllamaChatResponse = resp
        .json()
        .await
-        .map_err(|e| ServerFnError::new(format!("Failed to parse LiteLLM response: {e}")))?;
+        .map_err(|e| ServerFnError::new(format!("Failed to parse Ollama response: {e}")))?;

    body.choices
        .first()
        .map(|choice| choice.message.content.clone())
-        .ok_or_else(|| ServerFnError::new("Empty response from LiteLLM"))
+        .ok_or_else(|| ServerFnError::new("Empty response from Ollama"))
 }

 #[cfg(test)]
--- a/src/infrastructure/mod.rs
+++ b/src/infrastructure/mod.rs
@@ -3,8 +3,8 @@
 pub mod auth_check;
 pub mod chat;
 pub mod langgraph;
-pub mod litellm;
 pub mod llm;
+pub mod ollama;
 pub mod searxng;

 // Server-only modules (Axum handlers, state, configs, DB, etc.)
--- a/src/infrastructure/ollama.rs
+++ b/src/infrastructure/ollama.rs
@@ -0,0 +1,92 @@
+use dioxus::prelude::*;
+use serde::{Deserialize, Serialize};
+
+/// Status of a local Ollama instance, including connectivity and loaded models.
+///
+/// # Fields
+///
+/// * `online` - Whether the Ollama API responded successfully
+/// * `models` - List of model names currently available on the instance
+#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)]
+pub struct OllamaStatus {
+    pub online: bool,
+    pub models: Vec<String>,
+}
+
+/// Response from Ollama's `GET /api/tags` endpoint.
+#[cfg(feature = "server")]
+#[derive(Deserialize)]
+struct OllamaTagsResponse {
+    models: Vec<OllamaModel>,
+}
+
+/// A single model entry from Ollama's tags API.
+#[cfg(feature = "server")]
+#[derive(Deserialize)]
+struct OllamaModel {
+    name: String,
+}
+
+/// Check the status of a local Ollama instance by querying its tags endpoint.
+///
+/// Calls `GET <ollama_url>/api/tags` to list available models and determine
+/// whether the instance is reachable.
+///
+/// # Arguments
+///
+/// * `ollama_url` - Base URL of the Ollama instance (e.g. "http://localhost:11434")
+///
+/// # Returns
+///
+/// An `OllamaStatus` with `online: true` and model names if reachable,
+/// or `online: false` with an empty model list on failure
+///
+/// # Errors
+///
+/// Returns `ServerFnError` only on serialization issues; network failures
+/// are caught and returned as `online: false`
+#[post("/api/ollama-status")]
+pub async fn get_ollama_status(ollama_url: String) -> Result<OllamaStatus, ServerFnError> {
+    let state: crate::infrastructure::ServerState =
+        dioxus_fullstack::FullstackContext::extract().await?;
+
+    let base_url = if ollama_url.is_empty() {
+        state.services.ollama_url.clone()
+    } else {
+        ollama_url
+    };
+
+    let url = format!("{}/api/tags", base_url.trim_end_matches('/'));
+
+    let client = reqwest::Client::builder()
+        .timeout(std::time::Duration::from_secs(5))
+        .build()
+        .map_err(|e| ServerFnError::new(format!("HTTP client error: {e}")))?;
+
+    let resp = match client.get(&url).send().await {
+        Ok(r) if r.status().is_success() => r,
+        _ => {
+            return Ok(OllamaStatus {
+                online: false,
+                models: Vec::new(),
+            });
+        }
+    };
+
+    let body: OllamaTagsResponse = match resp.json().await {
+        Ok(b) => b,
+        Err(_) => {
+            return Ok(OllamaStatus {
+                online: true,
+                models: Vec::new(),
+            });
+        }
+    };
+
+    let models = body.models.into_iter().map(|m| m.name).collect();
+
+    Ok(OllamaStatus {
+        online: true,
+        models,
+    })
+}
--- a/src/infrastructure/provider_client.rs
+++ b/src/infrastructure/provider_client.rs
@@ -1,6 +1,6 @@
 //! Unified LLM provider dispatch.
 //!
-//! Routes chat completion requests to LiteLLM, OpenAI, Anthropic, or
+//! Routes chat completion requests to Ollama, OpenAI, Anthropic, or
 //! HuggingFace based on the session's provider setting. All providers
 //! except Anthropic use the OpenAI-compatible chat completions format.

@@ -20,11 +20,11 @@ pub struct ProviderMessage {
 ///
 /// # Arguments
 ///
-/// * `state` - Server state (for default LiteLLM URL/model)
-/// * `provider` - Provider name (`"litellm"`, `"openai"`, `"anthropic"`, `"huggingface"`)
+/// * `state` - Server state (for default Ollama URL/model)
+/// * `provider` - Provider name (`"ollama"`, `"openai"`, `"anthropic"`, `"huggingface"`)
 /// * `model` - Model ID
 /// * `messages` - Conversation history
-/// * `api_key` - API key (required for non-LiteLLM providers; LiteLLM uses server config)
+/// * `api_key` - API key (required for non-Ollama providers)
 /// * `stream` - Whether to request streaming
 ///
 /// # Returns
@@ -123,11 +123,11 @@ pub async fn send_chat_request(
                .send()
                .await
        }
-        // Default: LiteLLM proxy (OpenAI-compatible endpoint)
+        // Default: Ollama (OpenAI-compatible endpoint)
        _ => {
-            let base_url = &state.services.litellm_url;
+            let base_url = &state.services.ollama_url;
            let resolved_model = if model.is_empty() {
-                &state.services.litellm_model
+                &state.services.ollama_model
            } else {
                model
            };
@@ -137,15 +137,12 @@ pub async fn send_chat_request(
                "messages": messages,
                "stream": stream,
            });
-            let litellm_key = &state.services.litellm_api_key;
-            let mut request = client
+            client
                .post(&url)
                .header("content-type", "application/json")
-                .json(&body);
-            if !litellm_key.is_empty() {
-                request = request.header("Authorization", format!("Bearer {litellm_key}"));
-            }
-            request.send().await
+                .json(&body)
+                .send()
+                .await
        }
    }
 }
--- a/src/infrastructure/server_state.rs
+++ b/src/infrastructure/server_state.rs
@@ -45,7 +45,7 @@ pub struct ServerStateInner {
    pub keycloak: &'static KeycloakConfig,
    /// Outbound email settings.
    pub smtp: &'static SmtpConfig,
-    /// URLs for LiteLLM, SearXNG, LangChain, S3, etc.
+    /// URLs for Ollama, SearXNG, LangChain, S3, etc.
    pub services: &'static ServiceUrls,
    /// Stripe billing keys.
    pub stripe: &'static StripeConfig,
--- a/src/models/chat.rs
+++ b/src/models/chat.rs
@@ -60,8 +60,8 @@ pub struct Attachment {
 /// * `user_sub` - Keycloak subject ID (session owner)
 /// * `title` - Display title (auto-generated or user-renamed)
 /// * `namespace` - Grouping for sidebar sections
-/// * `provider` - LLM provider used (e.g. "litellm", "openai")
-/// * `model` - Model ID used (e.g. "qwen3-32b")
+/// * `provider` - LLM provider used (e.g. "ollama", "openai")
+/// * `model` - Model ID used (e.g. "llama3.1:8b")
 /// * `created_at` - ISO 8601 creation timestamp
 /// * `updated_at` - ISO 8601 last-activity timestamp
 /// * `article_url` - Source article URL (for News namespace sessions)
@@ -171,8 +171,8 @@ mod tests {
            user_sub: "user-1".into(),
            title: "Test Chat".into(),
            namespace: ChatNamespace::General,
-            provider: "litellm".into(),
-            model: "qwen3-32b".into(),
+            provider: "ollama".into(),
+            model: "llama3.1:8b".into(),
            created_at: "2025-01-01T00:00:00Z".into(),
            updated_at: "2025-01-01T01:00:00Z".into(),
            article_url: None,
@@ -189,7 +189,7 @@ mod tests {
            "_id": "mongo-id",
            "user_sub": "u1",
            "title": "t",
-            "provider": "litellm",
+            "provider": "ollama",
            "model": "m",
            "created_at": "2025-01-01",
            "updated_at": "2025-01-01"
@@ -205,7 +205,7 @@ mod tests {
            user_sub: "u1".into(),
            title: "t".into(),
            namespace: ChatNamespace::default(),
-            provider: "litellm".into(),
+            provider: "ollama".into(),
            model: "m".into(),
            created_at: "2025-01-01".into(),
            updated_at: "2025-01-01".into(),
@@ -223,7 +223,7 @@ mod tests {
            user_sub: "u1".into(),
            title: "t".into(),
            namespace: ChatNamespace::default(),
-            provider: "litellm".into(),
+            provider: "ollama".into(),
            model: "m".into(),
            created_at: "2025-01-01".into(),
            updated_at: "2025-01-01".into(),
--- a/src/models/organization.rs
+++ b/src/models/organization.rs
@@ -83,42 +83,6 @@ pub struct BillingUsage {
    pub billing_cycle_end: String,
 }

-/// Aggregated token usage statistics from LiteLLM's spend tracking API.
-///
-/// # Fields
-///
-/// * `total_spend` - Total cost in USD across all models
-/// * `total_prompt_tokens` - Sum of prompt (input) tokens
-/// * `total_completion_tokens` - Sum of completion (output) tokens
-/// * `total_tokens` - Sum of all tokens (prompt + completion)
-/// * `model_breakdown` - Per-model usage breakdown
-#[derive(Debug, Clone, Default, PartialEq, Serialize, Deserialize)]
-pub struct LitellmUsageStats {
-    pub total_spend: f64,
-    pub total_prompt_tokens: u64,
-    pub total_completion_tokens: u64,
-    pub total_tokens: u64,
-    pub model_breakdown: Vec<ModelUsage>,
-}
-
-/// Token and spend usage for a single LLM model.
-///
-/// # Fields
-///
-/// * `model` - Model identifier (e.g. "gpt-4", "claude-3-opus")
-/// * `spend` - Cost in USD for this model
-/// * `prompt_tokens` - Prompt (input) tokens consumed
-/// * `completion_tokens` - Completion (output) tokens generated
-/// * `total_tokens` - Total tokens (prompt + completion)
-#[derive(Debug, Clone, Default, PartialEq, Serialize, Deserialize)]
-pub struct ModelUsage {
-    pub model: String,
-    pub spend: f64,
-    pub prompt_tokens: u64,
-    pub completion_tokens: u64,
-    pub total_tokens: u64,
-}
-
 /// Organisation-level settings stored in MongoDB.
 ///
 /// These complement Keycloak's Organizations feature with
@@ -270,82 +234,4 @@ mod tests {
        assert_eq!(record.seats_used, 0);
        assert_eq!(record.tokens_used, 0);
    }
-
-    #[test]
-    fn litellm_usage_stats_default() {
-        let stats = LitellmUsageStats::default();
-        assert_eq!(stats.total_spend, 0.0);
-        assert_eq!(stats.total_prompt_tokens, 0);
-        assert_eq!(stats.total_completion_tokens, 0);
-        assert_eq!(stats.total_tokens, 0);
-        assert!(stats.model_breakdown.is_empty());
-    }
-
-    #[test]
-    fn litellm_usage_stats_serde_round_trip() {
-        let stats = LitellmUsageStats {
-            total_spend: 12.34,
-            total_prompt_tokens: 50_000,
-            total_completion_tokens: 25_000,
-            total_tokens: 75_000,
-            model_breakdown: vec![
-                ModelUsage {
-                    model: "gpt-4".into(),
-                    spend: 10.0,
-                    prompt_tokens: 40_000,
-                    completion_tokens: 20_000,
-                    total_tokens: 60_000,
-                },
-                ModelUsage {
-                    model: "claude-3-opus".into(),
-                    spend: 2.34,
-                    prompt_tokens: 10_000,
-                    completion_tokens: 5_000,
-                    total_tokens: 15_000,
-                },
-            ],
-        };
-        let json = serde_json::to_string(&stats).expect("serialize LitellmUsageStats");
-        let back: LitellmUsageStats =
-            serde_json::from_str(&json).expect("deserialize LitellmUsageStats");
-        assert_eq!(stats, back);
-    }
-
-    #[test]
-    fn model_usage_default() {
-        let usage = ModelUsage::default();
-        assert_eq!(usage.model, "");
-        assert_eq!(usage.spend, 0.0);
-        assert_eq!(usage.prompt_tokens, 0);
-        assert_eq!(usage.completion_tokens, 0);
-        assert_eq!(usage.total_tokens, 0);
-    }
-
-    #[test]
-    fn model_usage_serde_round_trip() {
-        let usage = ModelUsage {
-            model: "gpt-4-turbo".into(),
-            spend: 5.67,
-            prompt_tokens: 30_000,
-            completion_tokens: 15_000,
-            total_tokens: 45_000,
-        };
-        let json = serde_json::to_string(&usage).expect("serialize ModelUsage");
-        let back: ModelUsage = serde_json::from_str(&json).expect("deserialize ModelUsage");
-        assert_eq!(usage, back);
-    }
-
-    #[test]
-    fn litellm_usage_stats_empty_breakdown_round_trip() {
-        let stats = LitellmUsageStats {
-            total_spend: 0.0,
-            total_prompt_tokens: 0,
-            total_completion_tokens: 0,
-            total_tokens: 0,
-            model_breakdown: Vec::new(),
-        };
-        let json = serde_json::to_string(&stats).expect("serialize empty stats");
-        let back: LitellmUsageStats = serde_json::from_str(&json).expect("deserialize empty stats");
-        assert_eq!(stats, back);
-    }
 }
--- a/src/models/provider.rs
+++ b/src/models/provider.rs
@@ -3,8 +3,8 @@ use serde::{Deserialize, Serialize};
 /// Supported LLM provider backends.
 #[derive(Debug, Clone, PartialEq, Serialize, Deserialize)]
 pub enum LlmProvider {
-    /// LiteLLM proxy for unified model access
-    LiteLlm,
+    /// Self-hosted models via Ollama
+    Ollama,
    /// Hugging Face Inference API
    HuggingFace,
    /// OpenAI-compatible endpoints
@@ -17,7 +17,7 @@ impl LlmProvider {
    /// Returns the display name for a provider.
    pub fn label(&self) -> &'static str {
        match self {
-            Self::LiteLlm => "LiteLLM",
+            Self::Ollama => "Ollama",
            Self::HuggingFace => "Hugging Face",
            Self::OpenAi => "OpenAI",
            Self::Anthropic => "Anthropic",
@@ -29,7 +29,7 @@ impl LlmProvider {
 ///
 /// # Fields
 ///
-/// * `id` - Unique model identifier (e.g. "qwen3-32b")
+/// * `id` - Unique model identifier (e.g. "llama3.1:8b")
 /// * `name` - Human-readable display name
 /// * `provider` - Which provider hosts this model
 /// * `context_window` - Maximum context length in tokens
@@ -79,8 +79,8 @@ mod tests {
    use pretty_assertions::assert_eq;

    #[test]
-    fn llm_provider_label_litellm() {
-        assert_eq!(LlmProvider::LiteLlm.label(), "LiteLLM");
+    fn llm_provider_label_ollama() {
+        assert_eq!(LlmProvider::Ollama.label(), "Ollama");
    }

    #[test]
@@ -101,7 +101,7 @@ mod tests {
    #[test]
    fn llm_provider_serde_round_trip() {
        for variant in [
-            LlmProvider::LiteLlm,
+            LlmProvider::Ollama,
            LlmProvider::HuggingFace,
            LlmProvider::OpenAi,
            LlmProvider::Anthropic,
@@ -117,10 +117,10 @@ mod tests {
    #[test]
    fn model_entry_serde_round_trip() {
        let entry = ModelEntry {
-            id: "qwen3-32b".into(),
-            name: "Qwen3 32B".into(),
-            provider: LlmProvider::LiteLlm,
-            context_window: 32,
+            id: "llama3.1:8b".into(),
+            name: "Llama 3.1 8B".into(),
+            provider: LlmProvider::Ollama,
+            context_window: 8192,
        };
        let json = serde_json::to_string(&entry).expect("serialize ModelEntry");
        let back: ModelEntry = serde_json::from_str(&json).expect("deserialize ModelEntry");
--- a/src/models/user.rs
+++ b/src/models/user.rs
@@ -35,12 +35,12 @@ pub struct AuthInfo {
 /// Per-user LLM provider configuration stored in MongoDB.
 ///
 /// Controls which provider and model the user's chat sessions default
-/// to, and stores API keys for non-LiteLLM providers.
+/// to, and stores API keys for non-Ollama providers.
 #[derive(Debug, Clone, Default, PartialEq, Serialize, Deserialize)]
 pub struct UserProviderConfig {
-    /// Default provider name (e.g. "litellm", "openai")
+    /// Default provider name (e.g. "ollama", "openai")
    pub default_provider: String,
-    /// Default model ID (e.g. "qwen3-32b", "gpt-4o")
+    /// Default model ID (e.g. "llama3.1:8b", "gpt-4o")
    pub default_model: String,
    /// OpenAI API key (empty if not configured)
    #[serde(default, skip_serializing_if = "Option::is_none")]
@@ -51,8 +51,8 @@ pub struct UserProviderConfig {
    /// HuggingFace API key
    #[serde(default, skip_serializing_if = "Option::is_none")]
    pub huggingface_api_key: Option<String>,
-    /// Custom LiteLLM URL override (empty = use server default)
-    pub litellm_url_override: String,
+    /// Custom Ollama URL override (empty = use server default)
+    pub ollama_url_override: String,
 }

 /// Per-user preferences stored in MongoDB.
@@ -66,10 +66,10 @@ pub struct UserPreferences {
    pub org_id: String,
    /// User-selected news/search topics
    pub custom_topics: Vec<String>,
-    /// Per-user LiteLLM URL override (empty = use server default)
-    pub litellm_url_override: String,
-    /// Per-user LiteLLM model override (empty = use server default)
-    pub litellm_model_override: String,
+    /// Per-user Ollama URL override (empty = use server default)
+    pub ollama_url_override: String,
+    /// Per-user Ollama model override (empty = use server default)
+    pub ollama_model_override: String,
    /// Recently searched queries for quick access
    pub recent_searches: Vec<String>,
    /// LLM provider configuration
@@ -132,12 +132,12 @@ mod tests {
    #[test]
    fn user_provider_config_optional_keys_skip_none() {
        let cfg = UserProviderConfig {
-            default_provider: "litellm".into(),
-            default_model: "qwen3-32b".into(),
+            default_provider: "ollama".into(),
+            default_model: "llama3.1:8b".into(),
            openai_api_key: None,
            anthropic_api_key: None,
            huggingface_api_key: None,
-            litellm_url_override: String::new(),
+            ollama_url_override: String::new(),
        };
        let json = serde_json::to_string(&cfg).expect("serialize UserProviderConfig");
        assert!(!json.contains("openai_api_key"));
@@ -153,7 +153,7 @@ mod tests {
            openai_api_key: Some("sk-test".into()),
            anthropic_api_key: Some("ak-test".into()),
            huggingface_api_key: None,
-            litellm_url_override: "http://custom:4000".into(),
+            ollama_url_override: "http://custom:11434".into(),
        };
        let json = serde_json::to_string(&cfg).expect("serialize");
        let back: UserProviderConfig = serde_json::from_str(&json).expect("deserialize");
--- a/src/pages/dashboard.rs
+++ b/src/pages/dashboard.rs
@@ -25,8 +25,8 @@ const DEFAULT_TOPICS: &[&str] = &[
 ///
 /// State is persisted across sessions using localStorage:
 /// - `certifai_topics`: custom user-defined search topics
-/// - `certifai_litellm_url`: LiteLLM proxy URL for summarization
-/// - `certifai_litellm_model`: LiteLLM model ID for summarization
+/// - `certifai_ollama_url`: Ollama instance URL for summarization
+/// - `certifai_ollama_model`: Ollama model ID for summarization
 #[component]
 pub fn DashboardPage() -> Element {
    let locale = use_context::<Signal<Locale>>();
@@ -34,11 +34,11 @@ pub fn DashboardPage() -> Element {

    // Persistent state stored in localStorage
    let mut custom_topics = use_persistent("certifai_topics".to_string(), Vec::<String>::new);
-    // Default to empty so the server functions use LITELLM_URL / LITELLM_MODEL
+    // Default to empty so the server functions use OLLAMA_URL / OLLAMA_MODEL
    // from .env. Only stores a non-empty value when the user explicitly saves
    // an override via the Settings panel.
-    let mut litellm_url = use_persistent("certifai_litellm_url".to_string(), String::new);
-    let mut litellm_model = use_persistent("certifai_litellm_model".to_string(), String::new);
+    let mut ollama_url = use_persistent("certifai_ollama_url".to_string(), String::new);
+    let mut ollama_model = use_persistent("certifai_ollama_model".to_string(), String::new);

    // Reactive signals for UI state
    let mut active_topic = use_signal(|| "AI".to_string());
@@ -235,8 +235,8 @@ pub fn DashboardPage() -> Element {
                    onclick: move |_| {
                        let currently_shown = *show_settings.read();
                        if !currently_shown {
-                            settings_url.set(litellm_url.read().clone());
-                            settings_model.set(litellm_model.read().clone());
+                            settings_url.set(ollama_url.read().clone());
+                            settings_model.set(ollama_model.read().clone());
                        }
                        show_settings.set(!currently_shown);
                    },
@@ -247,16 +247,16 @@ pub fn DashboardPage() -> Element {
            // Settings panel (collapsible)
            if *show_settings.read() {
                div { class: "settings-panel",
-                    h4 { class: "settings-panel-title", "{t(l, \"dashboard.litellm_settings\")}" }
+                    h4 { class: "settings-panel-title", "{t(l, \"dashboard.ollama_settings\")}" }
                    p { class: "settings-hint",
                        "{t(l, \"dashboard.settings_hint\")}"
                    }
                    div { class: "settings-field",
-                        label { "{t(l, \"dashboard.litellm_url\")}" }
+                        label { "{t(l, \"dashboard.ollama_url\")}" }
                        input {
                            class: "settings-input",
                            r#type: "text",
-                            placeholder: "{t(l, \"dashboard.litellm_url_placeholder\")}",
+                            placeholder: "{t(l, \"dashboard.ollama_url_placeholder\")}",
                            value: "{settings_url}",
                            oninput: move |e| settings_url.set(e.value()),
                        }
@@ -274,8 +274,8 @@ pub fn DashboardPage() -> Element {
                    button {
                        class: "btn btn-primary",
                        onclick: move |_| {
-                            *litellm_url.write() = settings_url.read().trim().to_string();
-                            *litellm_model.write() = settings_model.read().trim().to_string();
+                            *ollama_url.write() = settings_url.read().trim().to_string();
+                            *ollama_model.write() = settings_model.read().trim().to_string();
                            show_settings.set(false);
                        },
                        "{t(l, \"common.save\")}"
@@ -320,14 +320,14 @@ pub fn DashboardPage() -> Element {
                                            news_session_id.set(None);


-                                            let ll_url = litellm_url.read().clone();
-                                            let mdl = litellm_model.read().clone();
+                                            let oll_url = ollama_url.read().clone();
+                                            let mdl = ollama_model.read().clone();
                                            spawn(async move {
                                                is_summarizing.set(true);
                                                match crate::infrastructure::llm::summarize_article(
                                                        snippet.clone(),
                                                        article_url,
-                                                        ll_url,
+                                                        oll_url,
                                                        mdl,
                                                    )
                                                    .await
@@ -373,8 +373,8 @@ pub fn DashboardPage() -> Element {
                            chat_messages: chat_messages.read().clone(),
                            is_chatting: *is_chatting.read(),
                            on_chat_send: move |question: String| {
-                                let ll_url = litellm_url.read().clone();
-                                let mdl = litellm_model.read().clone();
+                                let oll_url = ollama_url.read().clone();
+                                let mdl = ollama_model.read().clone();
                                let ctx = article_context.read().clone();
                                // Capture article info for News session creation
                                let card_title = selected_card
@@ -394,7 +394,7 @@ pub fn DashboardPage() -> Element {
                                    content: question.clone(),
                                });

-                                // Build full message history for LiteLLM
+                                // Build full message history for Ollama
                                let system_msg = format!(
                                    "You are a helpful assistant. The user is reading \
                                     a news article. Use the following context to answer \
@@ -422,7 +422,7 @@ pub fn DashboardPage() -> Element {
                                        match create_chat_session(
                                            card_title,
                                            "News".to_string(),
-                                            "litellm".to_string(),
+                                            "ollama".to_string(),
                                            mdl.clone(),
                                            card_url,
                                        )
@@ -458,7 +458,7 @@ pub fn DashboardPage() -> Element {
                                    }

                                    match crate::infrastructure::llm::chat_followup(
-                                        msgs, ll_url, mdl,
+                                        msgs, oll_url, mdl,
                                    )
                                    .await
                                    {
@@ -495,7 +495,7 @@ pub fn DashboardPage() -> Element {
                // Right: sidebar (when no card selected)
                if !has_selection {
                    DashboardSidebar {
-                        litellm_url: litellm_url.read().clone(),
+                        ollama_url: ollama_url.read().clone(),
                        trending: trending_topics.clone(),
                        recent_searches: recent_searches.read().clone(),
                        on_topic_click: move |topic: String| {
--- a/src/pages/organization/dashboard.rs
+++ b/src/pages/organization/dashboard.rs
@@ -2,14 +2,12 @@ use dioxus::prelude::*;

 use crate::components::{MemberRow, PageHeader};
 use crate::i18n::{t, tw, Locale};
-use crate::infrastructure::litellm::get_litellm_usage;
-use crate::models::{BillingUsage, LitellmUsageStats, MemberRole, OrgMember};
+use crate::models::{BillingUsage, MemberRole, OrgMember};

 /// Organization dashboard with billing stats, member table, and invite modal.
 ///
-/// Shows current billing usage (fetched from LiteLLM), a per-model
-/// breakdown table, a table of organization members with role
-/// management, and a button to invite new members.
+/// Shows current billing usage, a table of organization members
+/// with role management, and a button to invite new members.
 #[component]
 pub fn OrgDashboardPage() -> Element {
    let locale = use_context::<Signal<Locale>>();
@@ -22,20 +20,6 @@ pub fn OrgDashboardPage() -> Element {

    let members_list = members.read().clone();

-    // Compute date range: 1st of current month to today
-    let (start_date, end_date) = current_month_range();
-
-    // Fetch real usage stats from LiteLLM via server function.
-    // use_resource memoises and won't re-fire on parent re-renders.
-    let usage_resource = use_resource(move || {
-        let start = start_date.clone();
-        let end = end_date.clone();
-        async move { get_litellm_usage(start, end).await }
-    });
-
-    // Clone out of Signal to avoid holding the borrow across rsx!
-    let usage_snapshot = usage_resource.read().clone();
-
    // Format token counts for display
    let tokens_display = format_tokens(usage.tokens_used);
    let tokens_limit_display = format_tokens(usage.tokens_limit);
@@ -46,39 +30,26 @@ pub fn OrgDashboardPage() -> Element {
                title: t(l, "org.title"),
                subtitle: t(l, "org.subtitle"),
                actions: rsx! {
-                    button {
-                        class: "btn-primary",
-                        onclick: move |_| show_invite.set(true),
-                        {t(l, "org.invite_member")}
-                    }
+                    button { class: "btn-primary", onclick: move |_| show_invite.set(true), {t(l, "org.invite_member")} }
                },
            }

            // Stats bar
            div { class: "org-stats-bar",
                div { class: "org-stat",
-                    span { class: "org-stat-value",
-                        "{usage.seats_used}/{usage.seats_total}"
-                    }
+                    span { class: "org-stat-value", "{usage.seats_used}/{usage.seats_total}" }
                    span { class: "org-stat-label", {t(l, "org.seats_used")} }
                }
                div { class: "org-stat",
                    span { class: "org-stat-value", "{tokens_display}" }
-                    span { class: "org-stat-label",
-                        {tw(l, "org.of_tokens", &[("limit", &tokens_limit_display)])}
-                    }
+                    span { class: "org-stat-label", {tw(l, "org.of_tokens", &[("limit", &tokens_limit_display)])} }
                }
                div { class: "org-stat",
-                    span { class: "org-stat-value",
-                        "{usage.billing_cycle_end}"
-                    }
+                    span { class: "org-stat-value", "{usage.billing_cycle_end}" }
                    span { class: "org-stat-label", {t(l, "org.cycle_ends")} }
                }
            }

-            // LiteLLM usage stats section
-            {render_usage_section(l, &usage_snapshot)}
-
            // Members table
            div { class: "org-table-wrapper",
                table { class: "org-table",
@@ -143,144 +114,6 @@ pub fn OrgDashboardPage() -> Element {
    }
 }

-/// Render the LiteLLM usage stats section: totals bar + per-model table.
-///
-/// Shows a loading state while the resource is pending, an error/empty
-/// message on failure, and the full breakdown on success.
-fn render_usage_section(
-    l: Locale,
-    snapshot: &Option<Result<LitellmUsageStats, ServerFnError>>,
-) -> Element {
-    match snapshot {
-        None => rsx! {
-            div { class: "org-usage-loading",
-                span { {t(l, "org.loading_usage")} }
-            }
-        },
-        Some(Err(_)) => rsx! {
-            div { class: "org-usage-unavailable",
-                span { {t(l, "org.usage_unavailable")} }
-            }
-        },
-        Some(Ok(stats)) if stats.total_tokens == 0 && stats.model_breakdown.is_empty() => {
-            rsx! {
-                div { class: "org-usage-unavailable",
-                    span { {t(l, "org.usage_unavailable")} }
-                }
-            }
-        }
-        Some(Ok(stats)) => {
-            let spend_display = format!("${:.2}", stats.total_spend);
-            let total_display = format_tokens(stats.total_tokens);
-            // Free-tier LiteLLM doesn't provide prompt/completion split
-            let has_token_split =
-                stats.total_prompt_tokens > 0 || stats.total_completion_tokens > 0;
-
-            rsx! {
-                // Usage totals bar
-                div { class: "org-stats-bar",
-                    div { class: "org-stat",
-                        span { class: "org-stat-value", "{spend_display}" }
-                        span { class: "org-stat-label",
-                            {t(l, "org.total_spend")}
-                        }
-                    }
-                    div { class: "org-stat",
-                        span { class: "org-stat-value",
-                            "{total_display}"
-                        }
-                        span { class: "org-stat-label",
-                            {t(l, "org.total_tokens")}
-                        }
-                    }
-                    // Only show prompt/completion split when available
-                    if has_token_split {
-                        div { class: "org-stat",
-                            span { class: "org-stat-value",
-                                {format_tokens(stats.total_prompt_tokens)}
-                            }
-                            span { class: "org-stat-label",
-                                {t(l, "org.prompt_tokens")}
-                            }
-                        }
-                        div { class: "org-stat",
-                            span { class: "org-stat-value",
-                                {format_tokens(stats.total_completion_tokens)}
-                            }
-                            span { class: "org-stat-label",
-                                {t(l, "org.completion_tokens")}
-                            }
-                        }
-                    }
-                }
-
-                // Per-model breakdown table
-                if !stats.model_breakdown.is_empty() {
-                    h3 { class: "org-section-title",
-                        {t(l, "org.model_usage")}
-                    }
-                    div { class: "org-table-wrapper",
-                        table { class: "org-table",
-                            thead {
-                                tr {
-                                    th { {t(l, "org.model")} }
-                                    th { {t(l, "org.tokens")} }
-                                    th { {t(l, "org.spend")} }
-                                }
-                            }
-                            tbody {
-                                for model in &stats.model_breakdown {
-                                    tr { key: "{model.model}",
-                                        td { "{model.model}" }
-                                        td {
-                                            {format_tokens(model.total_tokens)}
-                                        }
-                                        td {
-                                            {format!(
-                                                "${:.2}", model.spend
-                                            )}
-                                        }
-                                    }
-                                }
-                            }
-                        }
-                    }
-                }
-            }
-        }
-    }
-}
-
-/// Compute the date range for the current billing month.
-///
-/// Returns `(start_date, end_date)` as `YYYY-MM-DD` strings where
-/// start_date is the 1st of the current month and end_date is today.
-///
-/// On the web target this uses `js_sys::Date` to read the browser clock.
-/// On the server target (SSR) it falls back to `chrono::Utc::now()`.
-fn current_month_range() -> (String, String) {
-    #[cfg(feature = "web")]
-    {
-        // js_sys::Date accesses the browser's local clock in WASM.
-        let now = js_sys::Date::new_0();
-        let year = now.get_full_year();
-        // JS months are 0-indexed, so add 1 for calendar month
-        let month = now.get_month() + 1;
-        let day = now.get_date();
-        let start = format!("{year:04}-{month:02}-01");
-        let end = format!("{year:04}-{month:02}-{day:02}");
-        (start, end)
-    }
-    #[cfg(not(feature = "web"))]
-    {
-        use chrono::Datelike;
-        let today = chrono::Utc::now().date_naive();
-        let start = format!("{:04}-{:02}-01", today.year(), today.month());
-        let end = today.format("%Y-%m-%d").to_string();
-        (start, end)
-    }
-}
-
 /// Formats a token count into a human-readable string (e.g. "1.2M").
 fn format_tokens(count: u64) -> String {
    const M: u64 = 1_000_000;
--- a/src/pages/providers.rs
+++ b/src/pages/providers.rs
@@ -13,8 +13,8 @@ pub fn ProvidersPage() -> Element {
    let locale = use_context::<Signal<Locale>>();
    let l = *locale.read();

-    let mut selected_provider = use_signal(|| LlmProvider::LiteLlm);
-    let mut selected_model = use_signal(|| "qwen3-32b".to_string());
+    let mut selected_provider = use_signal(|| LlmProvider::Ollama);
+    let mut selected_model = use_signal(|| "llama3.1:8b".to_string());
    let mut selected_embedding = use_signal(|| "nomic-embed-text".to_string());
    let mut api_key = use_signal(String::new);
    let mut saved = use_signal(|| false);
@@ -59,12 +59,12 @@ pub fn ProvidersPage() -> Element {
                                    "Hugging Face" => LlmProvider::HuggingFace,
                                    "OpenAI" => LlmProvider::OpenAi,
                                    "Anthropic" => LlmProvider::Anthropic,
-                                    _ => LlmProvider::LiteLlm,
+                                    _ => LlmProvider::Ollama,
                                };
                                selected_provider.set(prov);
                                saved.set(false);
                            },
-                            option { value: "LiteLLM", "LiteLLM" }
+                            option { value: "Ollama", "Ollama" }
                            option { value: "Hugging Face", "Hugging Face" }
                            option { value: "OpenAI", "OpenAI" }
                            option { value: "Anthropic", "Anthropic" }
@@ -156,28 +156,22 @@ pub fn ProvidersPage() -> Element {
 fn mock_models() -> Vec<ModelEntry> {
    vec![
        ModelEntry {
-            id: "qwen3-32b".into(),
-            name: "Qwen3 32B".into(),
-            provider: LlmProvider::LiteLlm,
-            context_window: 32,
-        },
-        ModelEntry {
-            id: "llama-3.3-70b".into(),
-            name: "Llama 3.3 70B".into(),
-            provider: LlmProvider::LiteLlm,
+            id: "llama3.1:8b".into(),
+            name: "Llama 3.1 8B".into(),
+            provider: LlmProvider::Ollama,
            context_window: 128,
        },
        ModelEntry {
-            id: "mistral-small-24b".into(),
-            name: "Mistral Small 24B".into(),
-            provider: LlmProvider::LiteLlm,
-            context_window: 32,
+            id: "llama3.1:70b".into(),
+            name: "Llama 3.1 70B".into(),
+            provider: LlmProvider::Ollama,
+            context_window: 128,
        },
        ModelEntry {
-            id: "deepseek-r1-70b".into(),
-            name: "DeepSeek R1 70B".into(),
-            provider: LlmProvider::LiteLlm,
-            context_window: 64,
+            id: "mistral:7b".into(),
+            name: "Mistral 7B".into(),
+            provider: LlmProvider::Ollama,
+            context_window: 32,
        },
        ModelEntry {
            id: "meta-llama/Llama-3.1-8B".into(),
@@ -206,7 +200,7 @@ fn mock_embeddings() -> Vec<EmbeddingEntry> {
        EmbeddingEntry {
            id: "nomic-embed-text".into(),
            name: "Nomic Embed Text".into(),
-            provider: LlmProvider::LiteLlm,
+            provider: LlmProvider::Ollama,
            dimensions: 768,
        },
        EmbeddingEntry {
Author	SHA1	Message	Date
Sharang Parnerkar	70095734d0	feat(analytics): integrate Langfuse with Keycloak SSO Some checks failed CI / Format (push) Successful in 3s Details CI / Clippy (push) Successful in 2m49s Details CI / Security Audit (push) Has been skipped Details CI / Tests (push) Has been skipped Details CI / E2E Tests (push) Has been skipped Details CI / Deploy (push) Has been skipped Details CI / Format (pull_request) Successful in 3s Details CI / Security Audit (pull_request) Has been cancelled Details CI / Tests (pull_request) Has been cancelled Details CI / E2E Tests (pull_request) Has been cancelled Details CI / Deploy (pull_request) Has been cancelled Details CI / Clippy (pull_request) Has been cancelled Details Add certifai-langfuse OIDC client to Keycloak realm export and configure the Langfuse Docker service with Keycloak SSO env vars (shared realm, account linking, local auth disabled). Replace the iframe-based analytics page with an informational landing since cross-origin SSO breaks in iframes. Users open Langfuse in a new tab where the active Keycloak session authenticates them transparently. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-02-25 20:54:47 +01:00
Sharang Parnerkar	c165841766	feat(developer): replace placeholder letters with Bootstrap icons on agents page All checks were successful CI / Security Audit (push) Has been skipped Details CI / Tests (push) Has been skipped Details CI / E2E Tests (push) Has been skipped Details CI / Deploy (push) Has been skipped Details CI / Format (push) Successful in 3s Details CI / Clippy (push) Successful in 2m48s Details Use dioxus_free_icons Bootstrap icons: BsCpu for hero, BsBook for docs, BsLightningCharge for getting started, BsGithub for GitHub, BsCodeSquare for examples, BsBoxArrowUpRight for API reference. Also fix conditional serde import for server feature. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-02-25 20:37:58 +01:00
Sharang Parnerkar	b846699dcf	test(e2e): update developer agents page tests for informational landing Some checks failed CI / Format (push) Successful in 3s Details CI / Clippy (push) Failing after 1m47s Details CI / Security Audit (push) Has been skipped Details CI / Tests (push) Has been skipped Details CI / E2E Tests (push) Has been skipped Details CI / Deploy (push) Has been skipped Details Replace obsolete iframe/placeholder tests with new tests covering the hero section, connection status, quick-start card grid, disabled API Reference card, and running agents table section. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-02-25 20:35:44 +01:00
Sharang Parnerkar	91a4b6ab34	feat(developer): replace agents iframe with informational landing and live agent table Some checks failed CI / Format (push) Successful in 3s Details CI / Clippy (push) Failing after 1m45s Details CI / Security Audit (push) Has been skipped Details CI / Tests (push) Has been skipped Details CI / E2E Tests (push) Has been skipped Details CI / Deploy (push) Has been skipped Details LangGraph is API-only with no web UI, so the ToolEmbed iframe pattern doesn't work. Replace it with an informational landing page featuring a hero section, connection status indicator, quick-start card grid linking to docs/GitHub/examples, and a live table of registered agents fetched from the LangGraph POST /assistants/search endpoint. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-02-25 18:37:33 +01:00
Sharang Parnerkar	40afc88317	feat(developer): add hybrid iframe integration for developer tools All checks were successful CI / Format (push) Successful in 4s Details CI / Clippy (push) Successful in 2m47s Details CI / Security Audit (push) Has been skipped Details CI / Tests (push) Has been skipped Details CI / E2E Tests (push) Has been skipped Details CI / Deploy (push) Has been skipped Details Replace placeholder pages with ToolEmbed component that embeds LangGraph, LangFlow, and Langfuse in iframes when configured, or shows "Not Configured" placeholders when URLs are empty. Add ServiceUrlsContext for passing service URLs through Dioxus context. Add docker-compose services for local development: LangFlow, LangGraph (trial), Langfuse with full dependency stack (Postgres, ClickHouse, Redis, MinIO). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-02-25 17:49:56 +01:00
Sharang Parnerkar	2efec74eca	ci: add E2E test job with Playwright and service containers All checks were successful CI / Format (push) Successful in 3s Details CI / Deploy (push) Has been skipped Details CI / Deploy (pull_request) Has been skipped Details CI / Clippy (push) Successful in 2m56s Details CI / Security Audit (push) Has been skipped Details CI / Tests (push) Has been skipped Details CI / Format (pull_request) Successful in 3s Details CI / Clippy (pull_request) Successful in 2m45s Details CI / Security Audit (pull_request) Has been skipped Details CI / Tests (pull_request) Has been skipped Details CI / E2E Tests (push) Has been skipped Details CI / E2E Tests (pull_request) Has been skipped Details Run Playwright browser tests on main and PRs to main after quality checks pass. Spins up MongoDB and SearXNG as services, starts Keycloak manually post-checkout (needs realm-export.json from repo), builds and serves the Dioxus app, then runs the full E2E suite. Deploy now gates on both unit and E2E tests. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-02-25 10:27:19 +01:00
Sharang Parnerkar	0e3e1a707f	test: add Playwright E2E test suite (30 tests) Add browser-level end-to-end tests covering public pages, Keycloak OAuth authentication flow, dashboard interactions, providers config, developer section, organization pages, and sidebar navigation. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-02-25 10:27:19 +01:00
Sharang Parnerkar	d4274387a9	test: add infrastructure logic unit tests (37 new tests) Add Phase 2 test coverage for infrastructure modules: - state.rs: 6 tests (defaults, serde round-trips, UserState deref/clone) - provider_client.rs: 2 tests (ProviderMessage serde) - llm.rs: 12 tests (FollowUpMessage serde, joined_len, parse_article_html extraction with article/main/role=main tags, fallback, exclusions, truncation, fragment skipping) - chat.rs: 17 tests (doc_to_chat_session, doc_to_chat_message BSON conversion, resolve_provider_url for all providers) Refactor: extract parse_article_html from fetch_article_text for testability without HTTP. Refactor resolve_provider_url to accept explicit params instead of full ServerState, avoiding need for MongoDB in tests. Total test count: 129 (up from 92). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-02-25 10:27:19 +01:00
Sharang Parnerkar	6890ed9b42	test: add comprehensive unit test suite (~85 new tests) Add unit tests across all model and server infrastructure layers, increasing test count from 7 to 92. Covers serde round-trips, enum methods, defaults, config parsing, error mapping, PKCE crypto (with RFC 7636 test vector), OAuth store, and SearXNG ranking/dedup logic. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-02-25 10:27:19 +01:00
Sharang Parnerkar	e244711644	feat(librechat): add OIDC HTTP patch and prompt=none for seamless SSO Switch to host networking so LibreChat can reach Keycloak on localhost. Patch openidStrategy.js to allow HTTP OIDC issuers for local dev (openid-client v6 enforces HTTPS by default). Add support for OPENID_AUTH_EXTRA_PARAMS env var and set prompt=none for automatic SSO login when a Keycloak session exists. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-02-25 10:27:06 +01:00
Sharang Parnerkar	1f7748c5b4	fix(librechat): remove prompt=none for local dev compatibility prompt=none causes silent failure when no Keycloak session exists yet. Standard OIDC flow still provides seamless SSO when the user has an active Keycloak session from the dashboard. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-02-25 10:27:06 +01:00