solved the problems in hf token

.env.example

@@ -0,0 +1,19 @@
+# Copy to ".env" in this same folder (Automatic-post-agents/) — service.py loads it on startup.
+# https://huggingface.co/settings/tokens
+
+# Required
+HF_TOKEN=hf_replace_with_your_token
+# LiteLLM Hugging Face form: huggingface/<org>/<model> (see Hub Inference / provider badges).
+# WebWorld is NOT on Inference Providers unless you use AGENTS_LLM_BASE_URL to your own endpoint.
+AGENTS_LLM_MODEL=your-model
+
+# Optional
+AGENTS_LLM_TEMPERATURE=0.72
+# Only if you use a dedicated OpenAI-compatible endpoint (e.g. HF Inference Endpoint URL):
+# AGENTS_LLM_BASE_URL=https://xxxx.region.aws.endpoints.huggingface.cloud
+
+# Optional: return full tracebacks from POST /generate (do not use in production)
+# AGENTS_DEBUG=1
+
+# Optional: uvicorn reads PORT when you use python service.py; you can also pass --port on the CLI
+# PORT=9000

.gitignore

@@ -0,0 +1,5 @@
+.env
+__pycache__/
+*.pyc
+.venv/
+venv/

README.md

@@ -15,7 +15,7 @@ FastAPI app that exposes:
 - `GET /health` — liveness check
 - `POST /generate` — body: `topic`, optional `feedback`, `memory_context`, `tone_instruction`; returns `{"post": "..."}`
 
-Set **`HF_TOKEN`** (or **`HUGGINGFACE_HUB_TOKEN`**) as a [Space secret](https://huggingface.co/docs/hub/spaces-overview#managing-secrets) (required). Optionally set **`AGENTS_LLM_MODEL`** (default: `huggingface/Qwen/WebWorld-32B`), **`AGENTS_LLM_TEMPERATURE`**, or **`AGENTS_LLM_BASE_URL`** for an OpenAI-compatible endpoint (see `crew.py`).
+Set **`HF_TOKEN`** (or **`HUGGINGFACE_HUB_TOKEN`**) as a [Space secret](https://huggingface.co/docs/hub/spaces-overview#managing-secrets) (required). Set **`AGENTS_LLM_MODEL`** as a Space **variable** (required — e.g. `huggingface/Qwen/Qwen2.5-7B-Instruct`; see `.env.example`). Optionally set **`AGENTS_LLM_TEMPERATURE`** or **`AGENTS_LLM_BASE_URL`** for an OpenAI-compatible endpoint (see `crew.py`). Do **not** use WebWorld with the public HF router only — it is not supported there unless **`AGENTS_LLM_BASE_URL`** points at your own endpoint.
 
 **Port:** the container listens on `$PORT` when the platform sets it, otherwise **7860** (Hugging Face Spaces default).
 

crew.py

@@ -1,18 +1,26 @@
 """
 Crew definition for the agents service.
 
-LLM is provided via LiteLLM. Default Hugging Face Hub model id:
-    huggingface/Qwen/WebWorld-32B
+LLM is provided via LiteLLM against Hugging Face. Set the model id in the environment — see
+AGENTS_LLM_MODEL in .env.example (LiteLLM form: huggingface/<org>/<model>).
 
-If that route fails, LiteLLM supports a Hub inference-provider prefix (only if the model's Hub page lists the
-provider for this model — do not guess the provider):
-    huggingface/<provider>/Qwen/WebWorld-32B
-Example shape (valid only when listed on the model card): huggingface/together/Qwen/WebWorld-32B
+Pick a model that your Hugging Face account can run via Inference Providers (check the model's Hub page
+for Inference / provider badges), or set AGENTS_LLM_BASE_URL to your own OpenAI-compatible endpoint.
 
-Set HF_TOKEN (Hugging Face access token). Optional:
-    AGENTS_LLM_MODEL         default: huggingface/Qwen/WebWorld-32B
+Qwen/WebWorld-32B is not served on the public Inference Providers router; use it only with
+AGENTS_LLM_BASE_URL (e.g. your own HF Inference Endpoint or vLLM).
+
+If a plain huggingface/<org>/<model> call fails, LiteLLM supports:
+    huggingface/<provider>/<org>/<model>
+only when that provider is listed on the model card for this model (do not guess the provider).
+
+Required environment:
+    HF_TOKEN (or HUGGINGFACE_HUB_TOKEN)
+    AGENTS_LLM_MODEL
+
+Optional:
     AGENTS_LLM_TEMPERATURE   float, default 0.5
-    AGENTS_LLM_BASE_URL      optional OpenAI-compatible base URL (e.g. HF Inference Endpoint)
+    AGENTS_LLM_BASE_URL      OpenAI-compatible base URL (e.g. HF Inference Endpoint)
 """
 from __future__ import annotations
 
@@ -26,6 +34,8 @@ from crewai.project import CrewBase, agent, crew, task
 
 def _resolve_hf_token() -> str:
     token = (os.getenv("HF_TOKEN") or os.getenv("HUGGINGFACE_HUB_TOKEN") or "").strip()
+    if (token.startswith('"') and token.endswith('"')) or (token.startswith("'") and token.endswith("'")):
+        token = token[1:-1].strip()
     if not token:
         raise RuntimeError(
             "Missing Hugging Face token. Set HF_TOKEN (or HUGGINGFACE_HUB_TOKEN) in the "
@@ -34,14 +44,42 @@ def _resolve_hf_token() -> str:
     return token
 
 
+def _strip_optional_env_quotes(value: str) -> str:
+    v = value.strip()
+    if (v.startswith('"') and v.endswith('"')) or (v.startswith("'") and v.endswith("'")):
+        v = v[1:-1].strip()
+    return v
+
+
+def _resolve_llm_model() -> str:
+    raw = (os.getenv("AGENTS_LLM_MODEL") or "").strip()
+    raw = _strip_optional_env_quotes(raw)
+    if not raw:
+        raise RuntimeError(
+            "Missing AGENTS_LLM_MODEL. Set it in the environment (see .env.example), e.g. "
+            "huggingface/Qwen/Qwen2.5-7B-Instruct"
+        )
+    return raw
+
+
 def _build_llm() -> LLM:
-    model = (os.getenv("AGENTS_LLM_MODEL") or "huggingface/Qwen/WebWorld-32B").strip()
+    model = _resolve_llm_model()
     temperature = float(os.getenv("AGENTS_LLM_TEMPERATURE", "0.5"))
 
     hf_token = _resolve_hf_token()
     os.environ["HF_TOKEN"] = hf_token
+    os.environ["HUGGINGFACE_HUB_TOKEN"] = hf_token
 
     base_url = (os.getenv("AGENTS_LLM_BASE_URL") or "").strip().rstrip("/")
+    base_url = _strip_optional_env_quotes(base_url) if base_url else ""
+
+    if "webworld" in model.lower() and not base_url:
+        raise RuntimeError(
+            "AGENTS_LLM_MODEL is set to a WebWorld model, which Hugging Face Inference Providers "
+            "does not host. Set AGENTS_LLM_MODEL to a routed instruct model, or run WebWorld on your "
+            "own endpoint and set AGENTS_LLM_BASE_URL to that OpenAI-compatible base URL."
+        )
+
     if base_url:
         return LLM(
             model=model,

requirements.txt

@@ -2,9 +2,10 @@ fastapi
 uvicorn[standard]
 python-dotenv
 
-# LLM: set HF_TOKEN (or HUGGINGFACE_HUB_TOKEN). LiteLLM model ids use huggingface/<org>/<model> (see crew.py).
-# If plain huggingface/Qwen/WebWorld-32B fails, try huggingface/<provider>/Qwen/WebWorld-32B only when the Hub
-# model page lists that inference provider for this model (e.g. a listed provider, not guessed).
+# LLM: set HF_TOKEN (or HUGGINGFACE_HUB_TOKEN) and AGENTS_LLM_MODEL (required — no default in code).
+# LiteLLM model ids: huggingface/<org>/<model>. See crew.py and .env.example.
+# If plain huggingface/<org>/<model> fails, try huggingface/<provider>/<org>/<model> only when the Hub
+# model page lists that inference provider (do not guess the provider).
 crewai
 crewai-tools
 litellm

service.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+import logging
 import os
 from pathlib import Path
 
@@ -17,6 +18,8 @@ _backend_env = Path(__file__).resolve().parents[1] / "backend" / ".env"
 load_dotenv(_agents_env, override=False)
 load_dotenv(_backend_env, override=False)
 
+logger = logging.getLogger(__name__)
+
 app = FastAPI()
 
 
@@ -54,8 +57,9 @@ def generate(req: GenerateRequest):
         result = crew_instance.kickoff(inputs={"topic": prompt})
         return {"post": str(result)}
     except Exception as exc:
-        # Don't leak internal stack traces cross-service.
+        # Don't leak internal stack traces in the HTTP response by default.
         msg = str(exc) or "Agent generation failed"
+        logger.warning("POST /generate failed: %s", msg, exc_info=True)
         if os.getenv("AGENTS_DEBUG", "").strip() in ("1", "true", "yes"):
             raise
         raise HTTPException(500, msg[:500])