Merge pull request #29 from chintakjoshi/anthropic2nvidianim

switching from Anthropic SDK to Nvidia NIM

Author: chintakjoshi

README.md

@@ -22,7 +22,7 @@ Highlights:
 ## Tech Stack
 
 - Frontend: React + TypeScript + Vite + Tailwind CSS
-- Backend: Flask + SQLAlchemy + Alembic + Pydantic + Anthropic SDK
+- Backend: Flask + SQLAlchemy + Alembic + Pydantic + OpenAI SDK (for NVIDIA NIM)
 - Database: PostgreSQL
 
 ## Prerequisites
@@ -42,12 +42,12 @@ copy .env.example .env
 ```
 
 Required for AI replies:
-- Set `ANTHROPIC_API_KEY` in `backend/.env`
+- Set `NIM_API_KEY` in `backend/.env`
 
 Key backend env values:
 - `DATABASE_URL`
 - `CORS_ORIGINS`
-- `ANTHROPIC_MODEL`
+- `NIM_MODEL`
 - `WEB_SEARCH_MAX_RESULTS`
 
 ### Frontend env

backend/.env.example

@@ -3,14 +3,15 @@ FLASK_ENV=development
 FLASK_DEBUG=1
 CORS_ORIGINS=http://localhost:5173
 DATABASE_URL=postgresql+psycopg://postgres:postgres@localhost:5432/manuscriptly_writer
-ANTHROPIC_API_KEY=your_anthropic_api_key
-ANTHROPIC_MODEL=claude-haiku-4-5-20251001
-ANTHROPIC_MAX_TOKENS=4000
-ANTHROPIC_TEMPERATURE=0.4
-ANTHROPIC_MAX_TOOL_ITERATIONS=5
-ANTHROPIC_RETRY_MAX_ATTEMPTS=3
-ANTHROPIC_RETRY_BASE_DELAY_SECONDS=0.75
-ANTHROPIC_RETRY_MAX_DELAY_SECONDS=4.0
+NIM_API_KEY=your_nvidia_nim_api_key
+NIM_BASE_URL=https://integrate.api.nvidia.com/v1
+NIM_MODEL=openai/gpt-oss-120b
+AI_MAX_TOKENS=4000
+AI_TEMPERATURE=0.4
+AI_MAX_TOOL_ITERATIONS=5
+AI_RETRY_MAX_ATTEMPTS=3
+AI_RETRY_BASE_DELAY_SECONDS=0.75
+AI_RETRY_MAX_DELAY_SECONDS=4.0
 WEB_SEARCH_API_URL=https://api.duckduckgo.com/
 WEB_SEARCH_TIMEOUT_SECONDS=8.0
 WEB_SEARCH_MAX_RESULTS=5

backend/app/agent_tools/handlers.py

@@ -5,8 +5,7 @@
 from time import perf_counter
 from typing import Any
 
-from anthropic import Anthropic
-from anthropic import APIConnectionError, APIError, APITimeoutError
+from openai import APIConnectionError, APIError, APITimeoutError, OpenAI
 from sqlalchemy.orm import Session
 
 from app.agent_tools.schemas import CreateContentIdeaInput, ExecutePlanInput, UpdateContentPlanInput, WebSearchInput
@@ -28,6 +27,13 @@ class ToolValidationError(ToolHandlerError):
     pass
 
 
+def _get_ai_client() -> OpenAI | None:
+    api_key = Config.NIM_API_KEY.strip()
+    if not api_key or api_key == "your_nvidia_nim_api_key":
+        return None
+    return OpenAI(api_key=api_key, base_url=Config.NIM_BASE_URL)
+
+
 def handle_create_content_idea(payload: CreateContentIdeaInput) -> dict[str, Any]:
     return _run_tool("create_content_idea", payload, _create_content_idea)
 
@@ -260,8 +266,8 @@ def _generate_plan_with_ai(
     constraints: dict[str, Any] | None,
     user_context: dict[str, Any],
 ) -> dict[str, Any] | None:
-    api_key = Config.ANTHROPIC_API_KEY.strip()
-    if not api_key or api_key == "your_anthropic_api_key":
+    client = _get_ai_client()
+    if client is None:
         return None
 
     prompt = (
@@ -277,13 +283,14 @@ def _generate_plan_with_ai(
     )
 
     try:
-        client = Anthropic(api_key=api_key)
-        response = client.messages.create(
-            model=Config.ANTHROPIC_MODEL,
-            max_tokens=Config.ANTHROPIC_MAX_TOKENS,
-            temperature=Config.ANTHROPIC_TEMPERATURE,
-            system="You are a precise content strategist. Follow output format exactly.",
-            messages=[{"role": "user", "content": prompt}],
+        response = client.chat.completions.create(
+            model=Config.NIM_MODEL,
+            max_tokens=Config.AI_MAX_TOKENS,
+            temperature=Config.AI_TEMPERATURE,
+            messages=[
+                {"role": "system", "content": "You are a precise content strategist. Follow output format exactly."},
+                {"role": "user", "content": prompt},
+            ],
         )
         text = _extract_text_from_response(response)
         return _parse_json_from_text(text)
@@ -297,8 +304,8 @@ def _generate_blog_with_ai(
     output_format: str,
     user_context: dict[str, Any],
 ) -> dict[str, Any] | None:
-    api_key = Config.ANTHROPIC_API_KEY.strip()
-    if not api_key or api_key == "your_anthropic_api_key":
+    client = _get_ai_client()
+    if client is None:
         return None
 
     prompt = (
@@ -316,13 +323,14 @@ def _generate_blog_with_ai(
     )
 
     try:
-        client = Anthropic(api_key=api_key)
-        response = client.messages.create(
-            model=Config.ANTHROPIC_MODEL,
-            max_tokens=Config.ANTHROPIC_MAX_TOKENS,
-            temperature=Config.ANTHROPIC_TEMPERATURE,
-            system="You are a senior blog writer. Follow output format exactly.",
-            messages=[{"role": "user", "content": prompt}],
+        response = client.chat.completions.create(
+            model=Config.NIM_MODEL,
+            max_tokens=Config.AI_MAX_TOKENS,
+            temperature=Config.AI_TEMPERATURE,
+            messages=[
+                {"role": "system", "content": "You are a senior blog writer. Follow output format exactly."},
+                {"role": "user", "content": prompt},
+            ],
         )
         text = _extract_text_from_response(response)
         parsed = _coerce_blog_payload_from_text(text, plan)
@@ -343,7 +351,7 @@ def _generate_blog_with_ai(
 
 
 def _generate_blog_markdown_retry_with_ai(
-    client: Anthropic,
+    client: OpenAI,
     plan: ContentPlan,
     writing_instructions: str | None,
     user_context: dict[str, Any],
@@ -360,12 +368,14 @@ def _generate_blog_markdown_retry_with_ai(
         f"User context (JSON):\n{json.dumps(user_context, ensure_ascii=True)}"
     )
     try:
-        response = client.messages.create(
-            model=Config.ANTHROPIC_MODEL,
-            max_tokens=Config.ANTHROPIC_MAX_TOKENS,
-            temperature=Config.ANTHROPIC_TEMPERATURE,
-            system="You are a senior blog writer. Return markdown only.",
-            messages=[{"role": "user", "content": prompt}],
+        response = client.chat.completions.create(
+            model=Config.NIM_MODEL,
+            max_tokens=Config.AI_MAX_TOKENS,
+            temperature=Config.AI_TEMPERATURE,
+            messages=[
+                {"role": "system", "content": "You are a senior blog writer. Return markdown only."},
+                {"role": "user", "content": prompt},
+            ],
         )
         return _extract_text_from_response(response)
     except (APIError, APIConnectionError, APITimeoutError, ToolHandlerError):
@@ -436,6 +446,13 @@ def _generate_blog_fallback(
 
 
 def _extract_text_from_response(response: Any) -> str:
+    choices = getattr(response, "choices", None)
+    if isinstance(choices, list) and choices:
+        message = getattr(choices[0], "message", None)
+        text = (getattr(message, "content", None) or "").strip() if message is not None else ""
+        if text:
+            return text
+
     text_parts: list[str] = []
     for block in getattr(response, "content", []):
         if getattr(block, "type", None) == "text":

backend/app/agent_tools/registry.py

@@ -34,6 +34,16 @@ def to_anthropic_tool(self) -> dict[str, Any]:
             "input_schema": self.input_model.model_json_schema(),
         }
 
+    def to_openai_tool(self) -> dict[str, Any]:
+        return {
+            "type": "function",
+            "function": {
+                "name": self.name,
+                "description": self.description,
+                "parameters": self.input_model.model_json_schema(),
+            },
+        }
+
 
 class ToolRegistry:
     def __init__(self) -> None:
@@ -55,3 +65,6 @@ def list(self) -> list[ToolDefinition]:
 
     def list_anthropic_tools(self) -> list[dict[str, Any]]:
         return [tool.to_anthropic_tool() for tool in self._tools.values()]
+
+    def list_openai_tools(self) -> list[dict[str, Any]]:
+        return [tool.to_openai_tool() for tool in self._tools.values()]

backend/app/api/routes/agent.py

@@ -24,7 +24,7 @@
 @agent_bp.post("/chat")
 def chat_with_agent():
     """
-    Create user message and generate assistant reply via Anthropic
+    Create user message and generate assistant reply via AI provider
     ---
     tags:
       - Agent
@@ -48,11 +48,11 @@ def chat_with_agent():
         schema:
           $ref: '#/definitions/ErrorResponse'
       500:
-        description: Anthropic configuration missing.
+        description: AI provider configuration missing.
         schema:
           $ref: '#/definitions/ErrorResponse'
       502:
-        description: Anthropic completion failed.
+        description: AI provider completion failed.
         schema:
           $ref: '#/definitions/ErrorResponse'
     """
@@ -86,7 +86,7 @@ def chat_with_agent():
         sse_manager.publish("message.created", user_payload, session_id=session_id)
         sse_manager.publish(
             "agent.response.started",
-            {"conversation_id": session_id, "model": Config.ANTHROPIC_MODEL},
+            {"conversation_id": session_id, "model": Config.NIM_MODEL},
             session_id=session_id,
         )
 
@@ -141,7 +141,7 @@ def emit_agent_event(event_name: str, payload: dict) -> None:
             {
                 "conversation_id": session_id,
                 "assistant_message_id": assistant_payload["id"],
-                "model": Config.ANTHROPIC_MODEL,
+                "model": Config.NIM_MODEL,
             },
             session_id=session_id,
         )
@@ -151,7 +151,7 @@ def emit_agent_event(event_name: str, payload: dict) -> None:
                 {
                     "user_message": user_payload,
                     "assistant_message": assistant_payload,
-                    "model": Config.ANTHROPIC_MODEL,
+                    "model": Config.NIM_MODEL,
                 }
             ),
             201,

backend/app/api/swagger.py

@@ -259,7 +259,7 @@
             "properties": {
                 "user_message": {"$ref": "#/definitions/Message"},
                 "assistant_message": {"$ref": "#/definitions/Message"},
-                "model": {"type": "string", "example": "claude-haiku-4-5-20251001"},
+                "model": {"type": "string", "example": "openai/gpt-oss-120b"},
             },
         },
         "StreamTestRequest": {

backend/app/core/config.py

@@ -17,14 +17,33 @@ class Config:
         "DATABASE_URL",
         "postgresql+psycopg://postgres:postgres@localhost:5432/manuscriptly_writer",
     )
-    ANTHROPIC_API_KEY = os.getenv("ANTHROPIC_API_KEY", "")
-    ANTHROPIC_MODEL = os.getenv("ANTHROPIC_MODEL", "claude-haiku-4-5-20251001")
-    ANTHROPIC_MAX_TOKENS = int(os.getenv("ANTHROPIC_MAX_TOKENS", "4000"))
-    ANTHROPIC_TEMPERATURE = float(os.getenv("ANTHROPIC_TEMPERATURE", "0.4"))
-    ANTHROPIC_MAX_TOOL_ITERATIONS = int(os.getenv("ANTHROPIC_MAX_TOOL_ITERATIONS", "5"))
-    ANTHROPIC_RETRY_MAX_ATTEMPTS = int(os.getenv("ANTHROPIC_RETRY_MAX_ATTEMPTS", "3"))
-    ANTHROPIC_RETRY_BASE_DELAY_SECONDS = float(os.getenv("ANTHROPIC_RETRY_BASE_DELAY_SECONDS", "0.75"))
-    ANTHROPIC_RETRY_MAX_DELAY_SECONDS = float(os.getenv("ANTHROPIC_RETRY_MAX_DELAY_SECONDS", "4.0"))
+    NIM_API_KEY = os.getenv("NIM_API_KEY", os.getenv("ANTHROPIC_API_KEY", ""))
+    NIM_BASE_URL = os.getenv("NIM_BASE_URL", "https://integrate.api.nvidia.com/v1")
+    NIM_MODEL = os.getenv("NIM_MODEL", os.getenv("ANTHROPIC_MODEL", "openai/gpt-oss-120b"))
+    AI_MAX_TOKENS = int(os.getenv("AI_MAX_TOKENS", os.getenv("ANTHROPIC_MAX_TOKENS", "4000")))
+    AI_TEMPERATURE = float(os.getenv("AI_TEMPERATURE", os.getenv("ANTHROPIC_TEMPERATURE", "0.4")))
+    AI_MAX_TOOL_ITERATIONS = int(
+        os.getenv("AI_MAX_TOOL_ITERATIONS", os.getenv("ANTHROPIC_MAX_TOOL_ITERATIONS", "5"))
+    )
+    AI_RETRY_MAX_ATTEMPTS = int(
+        os.getenv("AI_RETRY_MAX_ATTEMPTS", os.getenv("ANTHROPIC_RETRY_MAX_ATTEMPTS", "3"))
+    )
+    AI_RETRY_BASE_DELAY_SECONDS = float(
+        os.getenv("AI_RETRY_BASE_DELAY_SECONDS", os.getenv("ANTHROPIC_RETRY_BASE_DELAY_SECONDS", "0.75"))
+    )
+    AI_RETRY_MAX_DELAY_SECONDS = float(
+        os.getenv("AI_RETRY_MAX_DELAY_SECONDS", os.getenv("ANTHROPIC_RETRY_MAX_DELAY_SECONDS", "4.0"))
+    )
+
+    # Backward-compatible aliases for existing call sites/tests.
+    ANTHROPIC_API_KEY = NIM_API_KEY
+    ANTHROPIC_MODEL = NIM_MODEL
+    ANTHROPIC_MAX_TOKENS = AI_MAX_TOKENS
+    ANTHROPIC_TEMPERATURE = AI_TEMPERATURE
+    ANTHROPIC_MAX_TOOL_ITERATIONS = AI_MAX_TOOL_ITERATIONS
+    ANTHROPIC_RETRY_MAX_ATTEMPTS = AI_RETRY_MAX_ATTEMPTS
+    ANTHROPIC_RETRY_BASE_DELAY_SECONDS = AI_RETRY_BASE_DELAY_SECONDS
+    ANTHROPIC_RETRY_MAX_DELAY_SECONDS = AI_RETRY_MAX_DELAY_SECONDS
     WEB_SEARCH_API_URL = os.getenv("WEB_SEARCH_API_URL", "https://api.duckduckgo.com/")
     WEB_SEARCH_TIMEOUT_SECONDS = float(os.getenv("WEB_SEARCH_TIMEOUT_SECONDS", "8.0"))
     WEB_SEARCH_MAX_RESULTS = int(os.getenv("WEB_SEARCH_MAX_RESULTS", "5"))