@danhachuel/thunderbolt 0.3.51 → 0.3.53

package/hermes_ui/media_generation.py

@@ -116,6 +116,17 @@ def _image_value(payload: Any) -> tuple[bytes | None, str]:
             url = str(first.get("url") or first.get("image_url") or "").strip()
             if url:
                 return None, url
+    output = payload.get("output")
+    if isinstance(output, list) and output:
+        output = output[0]
+    if isinstance(output, Mapping):
+        output = output.get("url") or output.get("image_url") or output.get("image")
+    if isinstance(output, str):
+        if output.startswith(("http://", "https://")):
+            return None, output
+        output_data = _decode_data(output)
+        if output_data:
+            return output_data, ""
     result = payload.get("result")
     if isinstance(result, Mapping):
         data = _decode_data(result.get("image") or result.get("b64_json") or result.get("data"))
@@ -127,6 +138,44 @@ def _image_value(payload: Any) -> tuple[bytes | None, str]:
     return None, ""
 
 
+def _image_request_id(payload: Any) -> str:
+    if not isinstance(payload, Mapping):
+        return ""
+    for key in ("id", "prediction_id", "request_id", "task_id", "job_id"):
+        value = str(payload.get(key) or "").strip()
+        if value:
+            return value
+    return ""
+
+
+def _poll_image(card: Mapping[str, Any], request_id: str, *, attempts: int = 24, interval_seconds: float = 5.0) -> tuple[bytes | None, str]:
+    """Resolve a Replicate image prediction without exposing its token."""
+    style = str(card.get("api_style") or media_provider_definition(card.get("provider")).api_style)
+    base = _base_url(card)
+    endpoint = str(card.get("status_endpoint") or "").strip().replace("{id}", request_id)
+    if not endpoint:
+        if style != "replicate":
+            return None, ""
+        endpoint = f"{base}/predictions/{request_id}"
+    for index in range(max(1, attempts)):
+        try:
+            response = requests.get(endpoint, headers=_headers(card), timeout=60)
+            if response.status_code >= 400:
+                raise MediaGenerationError(f"Consulta de imagem devolveu HTTP {response.status_code}.")
+            payload = response.json()
+        except requests.RequestException as exc:
+            raise MediaGenerationError(f"Falha ao consultar a imagem: {str(exc)[:180]}") from exc
+        image_bytes, url = _image_value(payload if isinstance(payload, Mapping) else {})
+        if image_bytes or url:
+            return image_bytes, url
+        status = str((payload or {}).get("status") or "").lower() if isinstance(payload, Mapping) else ""
+        if status in {"failed", "error", "cancelled", "canceled"}:
+            raise MediaGenerationError("O provider marcou a tarefa de imagem como falhada.")
+        if index + 1 < attempts:
+            time.sleep(max(0.2, interval_seconds))
+    raise MediaGenerationError("O provider de imagem não concluiu dentro do limite de polling.")
+
+
 def _download_or_write(image_bytes: bytes | None, url: str, destination: Path, card: Mapping[str, Any]) -> Path:
     if image_bytes is None and url:
         try:
@@ -151,6 +200,8 @@ def _image_endpoint(card: Mapping[str, Any]) -> str:
         return explicit
     if style in {"openai_compatible", "huggingface", "agnes", "kie"}:
         return f"{base}/images/generations"
+    if style == "replicate":
+        return f"{base}/predictions"
     if style == "cloudflare":
         account_id = str(card.get("account_id") or "").strip()
         if not account_id:
@@ -184,6 +235,9 @@ def _image_request(card: dict[str, Any], prompt: str) -> Any:
     if style == "dashscope":
         body = {"model": _model(card), "input": {"prompt": constrained_prompt}, "parameters": {"n": 1}}
         return requests.post(endpoint, headers=_headers(card), json=body, timeout=180)
+    if style == "replicate":
+        body = {"version": _model(card), "input": {"prompt": constrained_prompt}}
+        return requests.post(endpoint, headers=_headers(card), json=body, timeout=180)
     body = {"model": _model(card), "prompt": constrained_prompt, "n": 1, "response_format": "b64_json"}
     return requests.post(endpoint, headers=_headers(card), json=body, timeout=180)
 
@@ -237,6 +291,11 @@ def generate_image_for_card(
     except ProviderRoutingError as exc:
         raise MediaGenerationError(str(exc)) from exc
     image_bytes, url = _image_value(routed.payload)
+    if not image_bytes and not url:
+        request_id = _image_request_id(routed.payload)
+        style = str(routed.card.get("api_style") or media_provider_definition(routed.card.get("provider")).api_style)
+        if request_id and style == "replicate":
+            image_bytes, url = _poll_image(routed.card, request_id)
     return _download_or_write(image_bytes, url, destination, routed.card)
 
 
@@ -294,6 +353,8 @@ def _video_endpoint(card: Mapping[str, Any]) -> str:
         return f"{base}/{_model(card).lstrip('/')}"
     if style in {"openai_compatible", "agnes", "kie"}:
         return f"{base}/videos/generations"
+    if style == "replicate":
+        return f"{base}/predictions"
     if style == "dashscope":
         return f"{base}/services/aigc/video-generation/video-synthesis"
     raise MediaGenerationError(f"O provider {card.get('provider')} não tem endpoint de vídeo configurado.")
@@ -316,12 +377,21 @@ def _video_request(card: dict[str, Any], prompt: str, image_url: str = "") -> An
     if style == "fal_queue":
         body.pop("model", None)
         return requests.post(endpoint, headers=_headers(card, fal=True), json=body, timeout=180)
+    if style == "replicate":
+        image_input_key = str(card.get("image_input_key") or "image").strip() or "image"
+        input_payload = {"prompt": body["prompt"]}
+        if image_url:
+            input_payload[image_input_key] = image_url
+        return requests.post(endpoint, headers=_headers(card), json={"version": _model(card), "input": input_payload}, timeout=180)
     return requests.post(endpoint, headers=_headers(card), json=body, timeout=180)
 
 
 def _video_result(payload: Mapping[str, Any]) -> tuple[str, str]:
-    direct = str(payload.get("video_url") or payload.get("url") or payload.get("output") or "").strip()
-    if direct:
+    raw_output = payload.get("output")
+    if isinstance(raw_output, list) and raw_output:
+        raw_output = raw_output[0]
+    direct = str(payload.get("video_url") or payload.get("url") or raw_output or "").strip()
+    if direct.startswith(("http://", "https://")):
         return direct, ""
     result = payload.get("result")
     if isinstance(result, Mapping):
@@ -350,6 +420,8 @@ def _poll_video(card: Mapping[str, Any], request_id: str, *, attempts: int = 24,
         endpoint = explicit_status.replace("{id}", request_id)
     elif style == "fal_queue":
         endpoint = f"{base}/requests/{request_id}/status"
+    elif style == "replicate":
+        endpoint = f"{base}/predictions/{request_id}"
     else:
         endpoint = f"{base}/videos/{request_id}"
     headers = _headers(card, fal=style == "fal_queue")
@@ -366,7 +438,7 @@ def _poll_video(card: Mapping[str, Any], request_id: str, *, attempts: int = 24,
         if url:
             return url
         status = str((payload or {}).get("status") or "").lower() if isinstance(payload, Mapping) else ""
-        if status in {"failed", "error", "cancelled"}:
+        if status in {"failed", "error", "cancelled", "canceled"}:
             raise ProviderCallError("O provider marcou a tarefa de vídeo como falhada.", category="provider", retryable=False)
         if index + 1 < attempts:
             time.sleep(max(0.2, interval_seconds))

package/hermes_ui/media_providers.py

@@ -53,6 +53,16 @@ MEDIA_PROVIDER_CATALOG: tuple[MediaProviderDefinition, ...] = (
         api_style="openai_compatible",
         description="Gateway multimodal com endpoints compatíveis e catálogo próprio.",
     ),
+    MediaProviderDefinition(
+        "replicate",
+        "Replicate",
+        default_base_url="https://api.replicate.com/v1",
+        supports_image=True,
+        supports_video=True,
+        supports_text=False,
+        api_style="replicate",
+        description="Predictions assíncronas; o campo Modelo aceita model ou model:version conforme a Replicate.",
+    ),
     MediaProviderDefinition(
         "agnes",
         "Agnes AI",

package/hermes_ui/notifications.py

@@ -21,6 +21,8 @@ EVENT_CATALOG: tuple[dict[str, str], ...] = (
     {"code": "script_stage_completed", "category": "Pipeline", "label": "Etapa de roteiro concluída", "description": "Quando a etapa de roteiro de uma tarefa terminar."},
     {"code": "title_generation_completed", "category": "Pipeline", "label": "Títulos gerados", "description": "Quando o pacote de títulos terminar de ser gerado."},
     {"code": "thumbnail_generation_completed", "category": "Pipeline", "label": "Thumbnail gerada", "description": "Quando a imagem final da thumbnail for criada."},
+    {"code": "influencer_content_completed", "category": "AI Influencers", "label": "Conteúdo de Influencer concluído", "description": "Quando uma imagem ou vídeo de AI Influencers terminar de ser gerado."},
+    {"code": "influencer_content_failed", "category": "AI Influencers", "label": "Conteúdo de Influencer falhou", "description": "Quando uma imagem ou vídeo de AI Influencers terminar com erro."},
     {"code": "blueprint_completed", "category": "Pipeline", "label": "Blueprint criado ou importado", "description": "Quando um Blueprint for criado, importado ou guardado."},
     {"code": "branding_completed", "category": "Pipeline", "label": "Branding criado ou importado", "description": "Quando um Branding for criado, importado ou guardado."},
     {"code": "niche_analysis_completed", "category": "Pipeline", "label": "Análise de nicho concluída", "description": "Quando uma análise Kaggle ou Apify terminar com resultados."},

package/hermes_ui/storage.py

@@ -230,6 +230,12 @@ DEFAULTS: dict[str, Any] = {
         "pollinations_api_key": "",
         "pollinations_base_url": "",
         "pollinations_model_name": "",
+        "influencer_db_backend": "Supabase",
+        "influencer_supabase_url": "",
+        "influencer_supabase_key": "",
+        "influencer_supabase_bucket": "ai-influencers",
+        "influencer_sqlite_path": "storage/state/ai_influencers.db",
+        "influencer_schema_version": 1,
         "log_level": "DEBUG",
         "listen_host": "127.0.0.1",
         "listen_port": 8080,
@@ -383,7 +389,7 @@ def _migrate_settings(settings: Any) -> tuple[dict[str, Any], bool]:
 
 
 def ensure_storage() -> None:
-    for path in [STATE, BLUEPRINTS / "canais", BLUEPRINTS / "nichos", BLUEPRINTS / "importados", BLUEPRINTS / "brandings", TIKTOK_PROMPT_MASTERS, MEDIA_DOWNLOADS, STORAGE / "brand", STORAGE / "scripts", STORAGE / "thumbnails", STORAGE / "videos", STORAGE / "artifacts", STORAGE / "skills", STORAGE / "metadata_cleaner", STORAGE / "metadata_cleaner" / "outputs", STORAGE / "music", STORAGE / "voice_previews", STORAGE / "python_editor", NICHES_DATA]:
+    for path in [STATE, BLUEPRINTS / "canais", BLUEPRINTS / "nichos", BLUEPRINTS / "importados", BLUEPRINTS / "brandings", TIKTOK_PROMPT_MASTERS, MEDIA_DOWNLOADS, STORAGE / "brand", STORAGE / "scripts", STORAGE / "thumbnails", STORAGE / "videos", STORAGE / "artifacts", STORAGE / "python_editor", STORAGE / "influencers", STORAGE / "metadata_cleaner", STORAGE / "metadata_cleaner" / "outputs", STORAGE / "music", STORAGE / "voice_previews", STORAGE / "python_editor", NICHES_DATA]:
         path.mkdir(parents=True, exist_ok=True)
     seed_blueprints()
     seed_prompt_masters()

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@danhachuel/thunderbolt",
-  "version": "0.3.51",
+  "version": "0.3.53",
   "description": "Thunderbolt — interface local para operação de canais faceless e motor MoneyPrinterTurbo",
   "license": "MIT",
   "main": "scripts/cli.mjs",
@@ -10,6 +10,7 @@
   },
   "files": [
     "app/main.py",
+    "app/influencers_ui.py",
     "app/modules/**/*.py",
     "hermes_ui/*.py",
     "integrations/*.py",
@@ -22,6 +23,7 @@
     "seed/skills/*.md",
     "seed/skills/*.py",
     "seed/references/*.md",
+    "seed/references/*.sql",
     "README.md",
     "MANUAL-INSTALACAO.md",
     "THIRD-PARTY-NOTICES.md",

package/requirements.txt

@@ -16,3 +16,4 @@ ytmusicapi>=1.12,<2
 pushtunes>=2.15,<3
 matplotlib>=3.7,<4
 kagglehub>=0.3,<1
+supabase>=2.4,<3

package/seed/references/ai_influencers_schema.sql

@@ -0,0 +1,69 @@
+-- Thunderbolt AI Influencers schema
+-- Apply in Supabase SQL Editor. Storage bucket creation is optional and may be
+-- performed from the dashboard with the configured bucket name.
+
+create table if not exists public.influencers (
+  id text primary key,
+  name text not null,
+  bio text not null default '',
+  instagram_business_id text not null default '',
+  language text not null default '',
+  profile_json jsonb not null default '{}'::jsonb,
+  created_at timestamptz not null default now(),
+  updated_at timestamptz not null default now()
+);
+
+create table if not exists public.influencer_assets (
+  id text primary key,
+  influencer_id text not null references public.influencers(id) on delete cascade,
+  asset_type text not null check (asset_type in ('image', 'document')),
+  original_name text not null,
+  stored_path text not null,
+  public_url text not null default '',
+  mime_type text not null default 'application/octet-stream',
+  size_bytes bigint not null default 0,
+  sha256 text not null,
+  document_json jsonb,
+  created_at timestamptz not null default now(),
+  unique (influencer_id, sha256)
+);
+create index if not exists idx_influencer_assets_influencer on public.influencer_assets(influencer_id, created_at);
+
+create table if not exists public.influencer_weekly_plans (
+  id text primary key,
+  influencer_id text not null references public.influencers(id) on delete cascade,
+  week text not null,
+  plan text not null default '',
+  created_at timestamptz not null default now(),
+  updated_at timestamptz not null default now(),
+  unique (influencer_id, week)
+);
+
+create table if not exists public.influencer_content (
+  id text primary key,
+  influencer_id text not null references public.influencers(id) on delete cascade,
+  content_type text not null check (content_type in ('image', 'video')),
+  prompt text not null default '',
+  caption text not null default '',
+  provider text not null default '',
+  model text not null default '',
+  platform text not null default '',
+  state text not null default 'queued',
+  artifact_path text not null default '',
+  provider_request_id text not null default '',
+  error text not null default '',
+  metadata_json jsonb not null default '{}'::jsonb,
+  created_at timestamptz not null default now(),
+  updated_at timestamptz not null default now()
+);
+create index if not exists idx_influencer_content_influencer on public.influencer_content(influencer_id, created_at);
+create index if not exists idx_influencer_content_state on public.influencer_content(state, updated_at);
+
+alter table public.influencers enable row level security;
+alter table public.influencer_assets enable row level security;
+alter table public.influencer_weekly_plans enable row level security;
+alter table public.influencer_content enable row level security;
+
+-- Configure least-privilege policies for the key used by Thunderbolt in your
+-- deployment. Do not make these tables public unless that is intentional.
+-- Example for a trusted local service_role key is deliberately omitted here.

package/seed/references/guide-supabase.md

@@ -1,6 +1,6 @@
 # Guia para criar e configurar uma conta Supabase
 
-Este guia mostra como criar uma conta gratuita no Supabase e preparar as tabelas e o armazenamento necessários para utilizar a base de dados com automações n8n.
+Este guia mostra como criar uma conta gratuita no Supabase e preparar as tabelas e o armazenamento necessários para utilizar a base de dados com automações n8n e com o módulo AI Influencers do Thunderbolt.
 
 Comece por abrir o [Supabase](https://supabase.com/) e criar uma nova conta gratuita.
 
@@ -56,6 +56,14 @@ Depois, carregue a imagem de referência para o bucket usando o botão `Upload f
 
 ![Upload files](https://github.com/user-attachments/assets/6cc83262-b92c-4f74-86c3-a45c9c0456f8)
 
+## 3. AI Influencers no Thunderbolt
+
+Para usar **AI Influencers > Personagens** e **Geração de Conteúdo IA** com Supabase, aplique o ficheiro `seed/references/ai_influencers_schema.sql` no SQL Editor. Ele cria as tabelas `influencers`, `influencer_assets`, `influencer_weekly_plans` e `influencer_content`, além dos índices e da activação de RLS.
+
+No Thunderbolt, abra **Configurações > Configuração API > API Keys > Banco de Dados Influencers**, seleccione **Supabase** e preencha o **Supabase Project URL**, a **Supabase API key** e o bucket de Storage. Crie o bucket com o mesmo nome configurado, por defeito `ai-influencers`, e confirme as políticas RLS/Storage antes de guardar imagens ou documentos. A chave não deve ser colocada no GitHub, nos workflows JSON ou em screenshots.
+
+O Thunderbolt usa o Supabase como backend seleccionado, não como executor de n8n: os assets são enviados ao Storage, os metadados ficam nas tabelas e os estados de geração ficam em `influencer_content`. Para uma execução totalmente local, seleccione **SQLite** no mesmo expander; a alternativa cria `storage/state/ai_influencers.db` e não utiliza a conta Supabase.
+
 A conta Supabase está pronta para ser utilizada pela automação.
 
 Fonte original: [guide-supabase.md no GitHub](https://github.com/gyoridavid/ai_agents_az/blob/main/episode_8/guide-supabase.md)