{% extends "base.html" %}
{% block title %}Knowledge & Retrieval — SimpleAudit{% endblock %}
{% block content %}
<div class="max-w-none">
  <div class="mb-5">
    <h1 class="text-lg font-semibold text-gray-100">Knowledge &amp; Retrieval</h1>
    <p class="text-sm text-gray-500 mt-1">Instance-wide document retrieval settings. Superusers only.</p>
  </div>

  {% include "admin/_tabs.html" %}

  {% if settings_error %}
  <div class="mb-5 rounded-lg border border-amber-800/60 bg-amber-900/20 px-4 py-3 text-sm text-amber-300">
    Knowledge and retrieval settings are unavailable: {{ settings_error }}
  </div>
  {% endif %}

  <form method="post" class="space-y-5">
    {% csrf_token %}

    <fieldset class="space-y-4 rounded-xl border border-gray-800 bg-surface-raised p-4">
      <div class="space-y-4">
        <div>
          <legend class="px-2 text-xs font-semibold uppercase tracking-wide text-gray-300">Embedding</legend>
          <p class="mt-1 text-sm text-gray-500">Choose the one service that converts uploaded documents into searchable vectors. These settings affect indexing and re-indexing.</p>
        </div>
        <div class="grid grid-cols-1 gap-4 md:grid-cols-2">
          <label class="form-control"><span class="label-text text-xs text-gray-400">Embedding service</span>
            <select id="embedding-engine" name="RAG_EMBEDDING_ENGINE" class="select select-sm mt-1 w-full bg-surface-overlay text-gray-100">
              <option value="" {% if not embedding.RAG_EMBEDDING_ENGINE %}selected{% endif %}>Local</option>
              <option value="ollama" {% if embedding.RAG_EMBEDDING_ENGINE == 'ollama' %}selected{% endif %}>Ollama</option>
              <option value="openai" {% if embedding.RAG_EMBEDDING_ENGINE == 'openai' %}selected{% endif %}>OpenAI-compatible</option>
              <option value="azure_openai" {% if embedding.RAG_EMBEDDING_ENGINE == 'azure_openai' %}selected{% endif %}>Azure OpenAI</option>
              {% if embedding.RAG_EMBEDDING_ENGINE and embedding.RAG_EMBEDDING_ENGINE != 'ollama' and embedding.RAG_EMBEDDING_ENGINE != 'openai' and embedding.RAG_EMBEDDING_ENGINE != 'azure_openai' %}<option value="{{ embedding.RAG_EMBEDDING_ENGINE }}" selected>{{ embedding.RAG_EMBEDDING_ENGINE }} (current)</option>{% endif %}
            </select>
            <span class="label-text-alt mt-1 text-gray-600">Only the selected service’s connection fields are shown.</span>
          </label>
          <label class="form-control"><span class="label-text text-xs text-gray-400">Embedding model</span><input type="text" name="RAG_EMBEDDING_MODEL" value="{{ embedding.RAG_EMBEDDING_MODEL|default:'' }}" placeholder="sentence-transformers/all-MiniLM-L6-v2" class="input input-sm mt-1 w-full bg-surface-overlay text-gray-100"><span class="label-text-alt mt-1 text-gray-600">The model used to turn document text into vectors.</span></label>
          <label class="form-control"><span class="label-text text-xs text-gray-400">Batch size</span><input type="number" min="1" name="RAG_EMBEDDING_BATCH_SIZE" value="{{ embedding.RAG_EMBEDDING_BATCH_SIZE|default:'' }}" class="input input-sm mt-1 w-full bg-surface-overlay text-gray-100"></label>
          <label class="form-control"><span class="label-text text-xs text-gray-400">Concurrent requests</span><input type="number" min="1" name="RAG_EMBEDDING_CONCURRENT_REQUESTS" value="{{ embedding.RAG_EMBEDDING_CONCURRENT_REQUESTS|default:'' }}" class="input input-sm mt-1 w-full bg-surface-overlay text-gray-100"></label>
        </div>
        <div data-embedding-provider="ollama" class="card card-border border-gray-800/80 bg-surface-overlay {% if embedding.RAG_EMBEDDING_ENGINE != 'ollama' %}hidden{% endif %}">
          <div class="card-body gap-3 p-4"><h3 class="text-sm font-medium text-gray-200">Ollama connection</h3><label class="form-control"><span class="label-text text-xs text-gray-400">Ollama URL</span><input type="text" name="RAG_OLLAMA_BASE_URL" value="{{ embedding.ollama_config.url|default:'' }}" placeholder="http://ollama:11434" class="input input-sm mt-1 w-full bg-surface text-gray-100"></label><label class="form-control"><span class="label-text text-xs text-gray-400">API key <span class="text-gray-600">(optional)</span></span><input type="password" name="RAG_OLLAMA_API_KEY" value="{{ embedding.ollama_config.key|default:'' }}" placeholder="Leave blank to keep existing key" class="input input-sm mt-1 w-full bg-surface text-gray-100"></label></div>
        </div>
        <div data-embedding-provider="openai" class="card card-border border-gray-800/80 bg-surface-overlay {% if embedding.RAG_EMBEDDING_ENGINE != 'openai' %}hidden{% endif %}">
          <div class="card-body gap-3 p-4"><h3 class="text-sm font-medium text-gray-200">OpenAI-compatible connection</h3><label class="form-control"><span class="label-text text-xs text-gray-400">Base URL</span><input type="text" name="RAG_OPENAI_BASE_URL" value="{{ embedding.openai_config.url|default:'' }}" placeholder="https://api.openai.com/v1" class="input input-sm mt-1 w-full bg-surface text-gray-100"></label><label class="form-control"><span class="label-text text-xs text-gray-400">API key</span><input type="password" name="RAG_OPENAI_API_KEY" value="{{ embedding.openai_config.key|default:'' }}" placeholder="Leave blank to keep existing key" class="input input-sm mt-1 w-full bg-surface text-gray-100"></label></div>
        </div>
        <div data-embedding-provider="azure_openai" class="card card-border border-gray-800/80 bg-surface-overlay {% if embedding.RAG_EMBEDDING_ENGINE != 'azure_openai' %}hidden{% endif %}">
          <div class="card-body gap-3 p-4"><h3 class="text-sm font-medium text-gray-200">Azure OpenAI connection</h3><label class="form-control"><span class="label-text text-xs text-gray-400">Base URL</span><input type="text" name="RAG_AZURE_OPENAI_BASE_URL" value="{{ embedding.azure_openai_config.url|default:'' }}" placeholder="https://…" class="input input-sm mt-1 w-full bg-surface text-gray-100"></label><div class="grid grid-cols-1 gap-3 md:grid-cols-2"><label class="form-control"><span class="label-text text-xs text-gray-400">API key</span><input type="password" name="RAG_AZURE_OPENAI_API_KEY" value="{{ embedding.azure_openai_config.key|default:'' }}" placeholder="Leave blank to keep existing key" class="input input-sm mt-1 w-full bg-surface text-gray-100"></label><label class="form-control"><span class="label-text text-xs text-gray-400">API version</span><input type="text" name="RAG_AZURE_OPENAI_API_VERSION" value="{{ embedding.azure_openai_config.version|default:'' }}" placeholder="2024-02-15-preview" class="input input-sm mt-1 w-full bg-surface text-gray-100"></label></div></div>
        </div>
        <label class="flex cursor-pointer items-start gap-3 text-sm text-gray-300"><input type="checkbox" name="ENABLE_ASYNC_EMBEDDING" {% if embedding.ENABLE_ASYNC_EMBEDDING %}checked{% endif %} class="toggle toggle-sm mt-0.5"><span><span class="block font-medium text-gray-200">Enable asynchronous embedding</span><span class="block text-xs text-gray-500">Process multiple document chunks without blocking the request.</span></span></label>
        <div role="note" class="alert alert-info text-xs">Provider keys are write-only here. Leave a key blank to keep the existing key.</div>
        <div class="rounded-lg border border-amber-800/60 bg-amber-900/10 p-3">
          <div class="flex flex-wrap items-center justify-between gap-3">
            <div>
              <h3 class="text-sm font-medium text-gray-200">Existing knowledge indexes</h3>
              <p class="mt-1 text-xs text-gray-500">Changing the embedding model does not rebuild existing vectors automatically.</p>
              {% if last_reindex %}
              <p class="mt-2 text-xs {% if last_reindex.status == 'failed' %}text-red-300{% elif last_reindex.status == 'succeeded' %}text-emerald-300{% else %}text-amber-300{% endif %}">
                Last re-index: {{ last_reindex.status|title }} · {{ last_reindex.started_at|date:"Y-m-d H:i T" }} · {{ last_reindex.embedding_model|default:"default model" }}
                {% if last_reindex.total != None %} · {{ last_reindex.success|default:0 }}/{{ last_reindex.total }} knowledge bases{% endif %}
              </p>
              {% else %}
              <p class="mt-2 text-xs text-gray-500">No Studio-triggered re-index has been recorded.</p>
              {% endif %}
            </div>
            <button type="submit" name="action" value="reindex" class="btn btn-sm btn-warning" onclick="return confirm('Re-index all existing knowledge bases? This may take a while.');">Re-index existing knowledge</button>
          </div>
          {% if last_reindex.status == 'succeeded' and reindex_matches_current == False %}
          <p class="mt-2 text-xs text-amber-300">The embedding settings changed after this re-index. Run it again to rebuild vectors with the current model.</p>
          {% endif %}
          {% if last_reindex.status == 'failed' and last_reindex.error %}<p class="mt-2 text-xs text-red-300">{{ last_reindex.error }}</p>{% endif %}
        </div>
      </div>
    </fieldset>

    <fieldset class="bg-surface-raised rounded-xl border border-gray-800 p-4 space-y-4">
      <legend class="px-2 text-xs font-semibold text-gray-300 uppercase tracking-wide">Retrieval</legend>
      <p class="text-xs text-gray-500">Controls how document matches are selected and assembled when an agent searches knowledge.</p>
      <div class="grid grid-cols-1 md:grid-cols-3 gap-4">
        {% include "admin/_rag_number.html" with name="TOP_K" label="Top K (initial matches)" value=rag.TOP_K help="How many candidate chunks to retrieve before optional reranking." %}
        {% include "admin/_rag_number.html" with name="TOP_K_RERANKER" label="Reranker Top K" value=rag.TOP_K_RERANKER help="How many candidates are passed to the reranker." %}
        {% include "admin/_rag_number.html" with name="RELEVANCE_THRESHOLD" label="Minimum relevance" value=rag.RELEVANCE_THRESHOLD step="0.01" help="Discard matches below this similarity score." %}
        {% include "admin/_rag_number.html" with name="HYBRID_BM25_WEIGHT" label="Keyword-search weight" value=rag.HYBRID_BM25_WEIGHT step="0.01" help="How much keyword matching contributes when hybrid search is enabled." %}
        {% include "admin/_rag_text.html" with name="RAG_RERANKING_ENGINE" label="Reranking engine" value=rag.RAG_RERANKING_ENGINE placeholder="" help="The service used to reorder the initial matches." %}
        {% include "admin/_rag_text.html" with name="RAG_RERANKING_MODEL" label="Reranking model" value=rag.RAG_RERANKING_MODEL placeholder="" help="The model used to score and reorder matches." %}
      </div>
      <div class="grid grid-cols-1 md:grid-cols-2 gap-2 text-sm text-gray-300">
        {% include "admin/_rag_checkbox.html" with name="RAG_FULL_CONTEXT" label="Full context mode" checked=rag.RAG_FULL_CONTEXT help="Include all selected matching text instead of trimming to a compact context." %}
        {% include "admin/_rag_checkbox.html" with name="BYPASS_EMBEDDING_AND_RETRIEVAL" label="Bypass document search" checked=rag.BYPASS_EMBEDDING_AND_RETRIEVAL help="Disable retrieval and answer without searching knowledge." %}
        {% include "admin/_rag_checkbox.html" with name="ENABLE_RAG_HYBRID_SEARCH" label="Hybrid semantic + keyword search" checked=rag.ENABLE_RAG_HYBRID_SEARCH help="Combine vector similarity with BM25 keyword matching." %}
        {% include "admin/_rag_checkbox.html" with name="ENABLE_RAG_HYBRID_SEARCH_ENRICHED_TEXTS" label="Enrich keyword-search text" checked=rag.ENABLE_RAG_HYBRID_SEARCH_ENRICHED_TEXTS help="Use enriched document text when calculating keyword matches." %}
      </div>
      <label class="block text-xs text-gray-400">RAG template
        <textarea name="RAG_TEMPLATE" rows="5" class="mt-1 w-full px-3 py-2 rounded-lg bg-surface-overlay border border-gray-700 text-sm text-gray-100 font-mono">{{ rag.RAG_TEMPLATE|default:'' }}</textarea>
      </label>
    </fieldset>

    <fieldset class="bg-surface-raised rounded-xl border border-gray-800 p-4 space-y-4">
      <legend class="px-2 text-xs font-semibold text-gray-300 uppercase tracking-wide">Chunking and files</legend>
      <div class="grid grid-cols-1 md:grid-cols-3 gap-4">
        {% include "admin/_rag_text.html" with name="TEXT_SPLITTER" label="Text splitter" value=rag.TEXT_SPLITTER placeholder="character, token" %}
        {% include "admin/_rag_text.html" with name="RAG_TOKENIZER_MODEL" label="Tokenizer model" value=rag.RAG_TOKENIZER_MODEL placeholder="Optional" %}
        {% include "admin/_rag_number.html" with name="CHUNK_SIZE" label="Chunk size" value=rag.CHUNK_SIZE %}
        {% include "admin/_rag_number.html" with name="CHUNK_OVERLAP" label="Chunk overlap" value=rag.CHUNK_OVERLAP %}
        {% include "admin/_rag_number.html" with name="CHUNK_MIN_SIZE_TARGET" label="Minimum chunk size" value=rag.CHUNK_MIN_SIZE_TARGET %}
        {% include "admin/_rag_text.html" with name="CONTENT_EXTRACTION_ENGINE" label="Content extraction engine" value=rag.CONTENT_EXTRACTION_ENGINE placeholder="Default" %}
        {% include "admin/_rag_text.html" with name="PDF_LOADER_MODE" label="PDF loader mode" value=rag.PDF_LOADER_MODE placeholder="page" %}
        {% include "admin/_rag_text.html" with name="FILE_MAX_SIZE" label="Maximum file size" value=rag.FILE_MAX_SIZE placeholder="" %}
        {% include "admin/_rag_text.html" with name="FILE_MAX_COUNT" label="Maximum file count" value=rag.FILE_MAX_COUNT placeholder="" %}
      </div>
      <div class="grid grid-cols-1 md:grid-cols-2 gap-2 text-sm text-gray-300">
        {% include "admin/_rag_checkbox.html" with name="ENABLE_MARKDOWN_HEADER_TEXT_SPLITTER" label="Markdown header splitting" checked=rag.ENABLE_MARKDOWN_HEADER_TEXT_SPLITTER %}
        {% include "admin/_rag_checkbox.html" with name="PDF_EXTRACT_IMAGES" label="Extract images from PDFs" checked=rag.PDF_EXTRACT_IMAGES %}
      </div>
      <label class="block text-xs text-gray-400">Allowed file extensions (comma-separated)
        <input type="text" name="ALLOWED_FILE_EXTENSIONS" value="{{ rag.ALLOWED_FILE_EXTENSIONS|join:', ' }}" class="mt-1 w-full px-3 py-2 rounded-lg bg-surface-overlay border border-gray-700 text-sm text-gray-100" placeholder="pdf, md, txt, docx">
      </label>
    </fieldset>

    <div class="flex items-center gap-3">
      <input type="hidden" name="tab" value="knowledge">
      <button type="submit" class="px-4 py-2 rounded-lg bg-accent text-white text-sm font-medium hover:bg-accent-hover">Save settings</button>
      <span class="text-xs text-amber-400">Changing the embedding model requires re-indexing existing knowledge bases.</span>
    </div>
  </form>
</div>
<script>
  (() => {
    const engine = document.getElementById('embedding-engine');
    const providers = document.querySelectorAll('[data-embedding-provider]');
    if (!engine) return;
    const updateProvider = () => {
      providers.forEach((provider) => {
        const active = provider.dataset.embeddingProvider === engine.value;
        provider.classList.toggle('hidden', !active);
        provider.querySelectorAll('input').forEach((input) => { input.disabled = !active; });
      });
    };
    engine.addEventListener('change', updateProvider);
    updateProvider();
  })();
</script>
{% endblock %}