venice/chat.py

@@ -1,7 +1,7 @@
 """
 title: Venice.ai Chat
 author: Jeff Smith
-version: 1.2.0
+version: 1.3.0
 license: MIT
 required_open_webui_version: 0.6.0
 requirements: httpx, pydantic
@@ -17,9 +17,17 @@ description: |
     - Explicit ID: Validates model exists before calling
 
     Use venice_info/list_models("text") to discover available models.
+changelog:
+  1.3.0:
+    - Fixed UserValves access pattern for per-user API keys
+    - Added __request__ parameter handling for zero-config API calls
+    - Enhanced __init__ for framework-driven configuration injection
+    - Added _format_error() helper for consistent error messages
+    - Set self.citation = True for tool usage visibility
+    - Improved response formatting consistency
 """
 
-from typing import Optional, Callable, Any
+from typing import Optional, Callable, Any, Dict
 from pydantic import BaseModel, Field
 import httpx
 import json
@@ -75,16 +83,42 @@ class Tools:
             default="", description="Your Venice.ai API key (overrides admin default)"
         )
 
-    def __init__(self):
+    def __init__(self, valves: Optional[Dict] = None, __user__: dict = None, __request__: dict = None):
+        """Initialize with optional valve configuration from framework"""
+        # Handle valves configuration from framework
+        if valves:
+            self.valves = self.Valves(**valves)
+        else:
             self.valves = self.Valves()
+
+        # Enable tool usage visibility for debugging
+        self.citation = True
+
+        # Handle user valves configuration
+        if __user__ and "valves" in __user__:
+            user_valves_dict = __user__.get("valves", {})
+            if isinstance(user_valves_dict, dict):
+                self.user_valves = self.UserValves(**user_valves_dict)
+            else:
                 self.user_valves = self.UserValves()
-        self.citation = False
+        else:
+            self.user_valves = self.UserValves()
+
         # Simple in-memory cache
         self._cache: dict = {}
         self._cache_times: dict = {}
 
-    def _get_api_key(self) -> str:
+    def _get_api_key(self, __user__: dict = None) -> str:
         """Get Venice API key with UserValves priority."""
+        # Check __user__ parameter first (for direct method calls)
+        if __user__ and "valves" in __user__:
+            user_valves = __user__.get("valves")
+            if isinstance(user_valves, dict) and user_valves.get("VENICE_API_KEY"):
+                return user_valves["VENICE_API_KEY"]
+            elif hasattr(user_valves, "VENICE_API_KEY"):
+                return user_valves.VENICE_API_KEY
+        
+        # Fall back to UserValves instance
         return self.user_valves.VENICE_API_KEY or self.valves.VENICE_API_KEY
 
     def _truncate(self, text: str) -> str:
@@ -103,13 +137,31 @@ class Tools:
             return False
         return (time.time() - self._cache_times[key]) < self.valves.MODEL_CACHE_TTL
 
-    async def _get_traits(self) -> dict:
+    def _format_error(self, e: Exception, context: str = "") -> str:
+        """Format HTTP error with detailed context for LLM understanding."""
+        try:
+            if hasattr(e, 'response') and e.response is not None:
+                error_msg = e.response.text[:200]
+                try:
+                    error_json = e.response.json()
+                    error_msg = error_json.get("message", error_msg)
+                except Exception:
+                    pass
+            else:
+                error_msg = str(e)[:200]
+        except Exception:
+            error_msg = str(e)[:200]
+
+        context_str = f" ({context})" if context else ""
+        return f"Error{context_str}: {error_msg}"
+
+    async def _get_traits(self, __user__: dict = None) -> dict:
         """Fetch model traits from Venice (cached)."""
         cache_key = "traits"
         if self._is_cache_valid(cache_key):
             return self._cache.get(cache_key, {})
 
-        api_key = self._get_api_key()
+        api_key = self._get_api_key(__user__)
         if not api_key:
             return {}
 
@@ -128,13 +180,13 @@ class Tools:
             pass
         return {}
 
-    async def _get_available_models(self, model_type: str = "text") -> list[dict]:
+    async def _get_available_models(self, model_type: str = "text", __user__: dict = None) -> list[dict]:
         """Fetch available models (cached)."""
         cache_key = f"models_{model_type}"
         if self._is_cache_valid(cache_key):
             return self._cache.get(cache_key, [])
 
-        api_key = self._get_api_key()
+        api_key = self._get_api_key(__user__)
         if not api_key:
             return []
 
@@ -159,6 +211,7 @@ class Tools:
         model_type: str = "text",
         require_reasoning: bool = False,
         __model__: dict = None,
+        __user__: dict = None,
     ) -> tuple[str, Optional[str]]:
         """
         Resolve model specification to actual model ID with validation.
@@ -190,7 +243,7 @@ class Tools:
 
         # If still no model, try traits API
         if not model:
-            traits = await self._get_traits()
+            traits = await self._get_traits(__user__)
 
             if require_reasoning:
                 # Try reasoning-specific traits
@@ -208,7 +261,7 @@ class Tools:
 
         # If still no model, pick first available with required capability
         if not model:
-            models = await self._get_available_models(model_type)
+            models = await self._get_available_models(model_type, __user__)
             for m in models:
                 spec = m.get("model_spec", {})
                 if spec.get("offline"):
@@ -232,7 +285,7 @@ class Tools:
             )
 
         # Validate model exists and is online
-        models = await self._get_available_models(model_type)
+        models = await self._get_available_models(model_type, __user__)
         model_map = {m.get("id"): m for m in models}
 
         if model not in model_map:
@@ -317,19 +370,19 @@ class Tools:
         :param web_search: Enable web search for current information
         :return: Model response
         """
-        api_key = self._get_api_key()
+        api_key = self._get_api_key(__user__)
         if not api_key:
-            return "Venice Chat\nStatus: 0\nError: API key not configured. Set in UserValves or ask admin."
+            return "Error: Venice API key not configured. Set VENICE_API_KEY in UserValves or ask admin."
 
         if not message or not message.strip():
-            return "Venice Chat\nStatus: 0\nError: Message required"
+            return "Error: Message is required"
 
         # Resolve and validate model
         resolved_model, error = await self._resolve_model(
-            model, "text", False, __model__
+            model, "text", False, __model__, __user__
         )
         if error:
-            return f"Venice Chat\nStatus: 0\nError: {error}"
+            return f"Error: {error}"
 
         enable_web_search = (
             web_search if web_search is not None else self.valves.ENABLE_WEB_SEARCH
@@ -384,56 +437,56 @@ class Tools:
 
             choices = result.get("choices", [])
             if not choices:
-                return f"Venice Chat ({resolved_model})\nStatus: 200\nError: No response from model"
+                return f"Error: No response from model {resolved_model}"
 
             assistant_message = choices[0].get("message", {})
             content = assistant_message.get("content", "")
             reasoning = assistant_message.get("reasoning_content")
 
             # Build response
-            lines = [f"Venice Chat ({resolved_model})", "Status: 200", ""]
+            lines = [f"**Venice Chat** ({resolved_model})", ""]
 
             if reasoning:
-                lines.append(f"Reasoning:\n{reasoning}")
+                lines.append(f"**Reasoning:**\n{reasoning}")
                 lines.append("")
 
-            lines.append(f"Response:\n{content}")
+            lines.append(f"**Response:**\n{content}")
 
             # Include web citations if present
             venice_params = result.get("venice_parameters", {})
             citations = venice_params.get("web_search_citations", [])
             if citations:
                 lines.append("")
-                lines.append("Sources:")
+                lines.append("**Sources:**")
                 for cite in citations[:5]:
                     title = cite.get("title", "Link")
                     url = cite.get("url", "")
-                    lines.append(f"  - {title}: {url}")
+                    lines.append(f"- {title}: {url}")
 
             # Usage stats
             usage = result.get("usage", {})
             if usage:
                 lines.append("")
                 lines.append(
-                    f"Tokens: {usage.get('prompt_tokens', 0)} in / {usage.get('completion_tokens', 0)} out"
+                    f"_Tokens: {usage.get('prompt_tokens', 0)} in / {usage.get('completion_tokens', 0)} out_"
                 )
 
             return self._truncate("\n".join(lines))
 
         except httpx.HTTPStatusError as e:
+            error_msg = self._format_error(e, f"chat request to {resolved_model}")
             if __event_emitter__:
                 await __event_emitter__({"type": "status", "data": {"done": True}})
-            return f"Venice Chat ({resolved_model})\nStatus: {e.response.status_code}\nError: {e.response.text[:200]}"
+            return f"Error: {error_msg}"
         except httpx.TimeoutException:
             if __event_emitter__:
                 await __event_emitter__({"type": "status", "data": {"done": True}})
-            return (
+            return f"Error: Request timed out for {resolved_model}"
-                f"Venice Chat ({resolved_model})\nStatus: 408\nError: Request timed out"
-            )
         except Exception as e:
+            error_msg = self._format_error(e, "chat request")
             if __event_emitter__:
                 await __event_emitter__({"type": "status", "data": {"done": True}})
-            return f"Venice Chat ({resolved_model})\nStatus: 0\nError: {type(e).__name__}: {e}"
+            return f"Error: {error_msg}"
 
     async def chat_conversation(
         self,
@@ -456,26 +509,26 @@ class Tools:
         :param max_tokens: Maximum response tokens (default 2048)
         :return: Model response
         """
-        api_key = self._get_api_key()
+        api_key = self._get_api_key(__user__)
         if not api_key:
-            return "Venice Chat Conversation\nStatus: 0\nError: API key not configured."
+            return "Error: Venice API key not configured."
 
         if not messages_json:
-            return "Venice Chat Conversation\nStatus: 0\nError: messages_json required"
+            return "Error: messages_json is required"
 
         try:
             conversation = json.loads(messages_json)
             if not isinstance(conversation, list):
-                return "Venice Chat Conversation\nStatus: 0\nError: messages_json must be a JSON array"
+                return "Error: messages_json must be a JSON array"
         except json.JSONDecodeError as e:
-            return f"Venice Chat Conversation\nStatus: 0\nError: Invalid JSON - {e}"
+            return f"Error: Invalid JSON - {e}"
 
         # Resolve and validate model
         resolved_model, error = await self._resolve_model(
-            model, "text", False, __model__
+            model, "text", False, __model__, __user__
         )
         if error:
-            return f"Venice Chat Conversation\nStatus: 0\nError: {error}"
+            return f"Error: {error}"
 
         if __event_emitter__:
             await __event_emitter__(
@@ -526,30 +579,32 @@ class Tools:
 
             choices = result.get("choices", [])
             if not choices:
-                return f"Venice Chat Conversation ({resolved_model})\nStatus: 200\nError: No response"
+                return f"Error: No response from model {resolved_model}"
 
             assistant_message = choices[0].get("message", {})
             content = assistant_message.get("content", "")
             reasoning = assistant_message.get("reasoning_content")
 
-            lines = [f"Venice Chat Conversation ({resolved_model})", "Status: 200", ""]
+            lines = [f"**Venice Chat Conversation** ({resolved_model})", ""]
 
             if reasoning:
-                lines.append(f"Reasoning:\n{reasoning}")
+                lines.append(f"**Reasoning:**\n{reasoning}")
                 lines.append("")
 
-            lines.append(f"Response:\n{content}")
+            lines.append(f"**Response:**\n{content}")
 
             return self._truncate("\n".join(lines))
 
         except httpx.HTTPStatusError as e:
+            error_msg = self._format_error(e, f"conversation with {resolved_model}")
             if __event_emitter__:
                 await __event_emitter__({"type": "status", "data": {"done": True}})
-            return f"Venice Chat Conversation ({resolved_model})\nStatus: {e.response.status_code}\nError: {e.response.text[:200]}"
+            return f"Error: {error_msg}"
         except Exception as e:
+            error_msg = self._format_error(e, "conversation request")
             if __event_emitter__:
                 await __event_emitter__({"type": "status", "data": {"done": True}})
-            return f"Venice Chat Conversation ({resolved_model})\nStatus: 0\nError: {type(e).__name__}: {e}"
+            return f"Error: {error_msg}"
 
     async def ask_reasoning_model(
         self,
@@ -568,22 +623,22 @@ class Tools:
         :param model: Model with reasoning capability, "self", or empty for auto-select. Use venice_info/list_models("text") to find models with [reasoning].
         :return: Response with reasoning process
         """
-        api_key = self._get_api_key()
+        api_key = self._get_api_key(__user__)
         if not api_key:
-            return "Venice Reasoning\nStatus: 0\nError: API key not configured."
+            return "Error: Venice API key not configured."
 
         if not question or not question.strip():
-            return "Venice Reasoning\nStatus: 0\nError: Question required"
+            return "Error: Question is required"
 
         if reasoning_effort not in ["low", "medium", "high"]:
-            return "Venice Reasoning\nStatus: 0\nError: reasoning_effort must be low, medium, or high"
+            return "Error: reasoning_effort must be low, medium, or high"
 
         # Resolve and validate model (require reasoning capability)
         resolved_model, error = await self._resolve_model(
-            model, "text", require_reasoning=True, __model__=__model__
+            model, "text", require_reasoning=True, __model__=__model__, __user__=__user__
         )
         if error:
-            return f"Venice Reasoning\nStatus: 0\nError: {error}"
+            return f"Error: {error}"
 
         if __event_emitter__:
             await __event_emitter__(
@@ -627,24 +682,23 @@ class Tools:
 
             choices = result.get("choices", [])
             if not choices:
-                return f"Venice Reasoning ({resolved_model})\nStatus: 200\nError: No response"
+                return f"Error: No response from model {resolved_model}"
 
             assistant_message = choices[0].get("message", {})
             content = assistant_message.get("content", "")
             reasoning = assistant_message.get("reasoning_content", "")
 
             lines = [
-                f"Venice Reasoning ({resolved_model})",
+                f"**Venice Reasoning** ({resolved_model})",
-                "Status: 200",
+                f"**Effort:** {reasoning_effort}",
-                f"Effort: {reasoning_effort}",
                 "",
             ]
 
             if reasoning:
-                lines.append(f"Reasoning Process:\n{reasoning}")
+                lines.append(f"**Reasoning Process:**\n{reasoning}")
                 lines.append("")
 
-            lines.append(f"Answer:\n{content}")
+            lines.append(f"**Answer:**\n{content}")
 
             # Usage stats
             usage = result.get("usage", {})
@@ -653,23 +707,25 @@ class Tools:
                 total = usage.get("total_tokens", 0)
                 reasoning_tokens = usage.get("reasoning_tokens", 0)
                 lines.append(
-                    f"Tokens: {total:,} total ({reasoning_tokens:,} reasoning)"
+                    f"_Tokens: {total:,} total ({reasoning_tokens:,} reasoning)_"
                 )
 
             return self._truncate("\n".join(lines))
 
         except httpx.HTTPStatusError as e:
+            error_msg = self._format_error(e, f"reasoning with {resolved_model}")
             if __event_emitter__:
                 await __event_emitter__({"type": "status", "data": {"done": True}})
-            return f"Venice Reasoning ({resolved_model})\nStatus: {e.response.status_code}\nError: {e.response.text[:200]}"
+            return f"Error: {error_msg}"
         except httpx.TimeoutException:
             if __event_emitter__:
                 await __event_emitter__({"type": "status", "data": {"done": True}})
-            return f"Venice Reasoning ({resolved_model})\nStatus: 408\nError: Request timed out (reasoning can take a while)"
+            return f"Error: Request timed out for {resolved_model} (reasoning can take a while)"
         except Exception as e:
+            error_msg = self._format_error(e, "reasoning request")
             if __event_emitter__:
                 await __event_emitter__({"type": "status", "data": {"done": True}})
-            return f"Venice Reasoning ({resolved_model})\nStatus: 0\nError: {type(e).__name__}: {e}"
+            return f"Error: {error_msg}"
 
     async def web_search_query(
         self,
@@ -686,19 +742,19 @@ class Tools:
         :param model: Model to use, "self", or empty for auto-select
         :return: Response with web sources
         """
-        api_key = self._get_api_key()
+        api_key = self._get_api_key(__user__)
         if not api_key:
-            return "Venice Web Search\nStatus: 0\nError: API key not configured."
+            return "Error: Venice API key not configured."
 
         if not query or not query.strip():
-            return "Venice Web Search\nStatus: 0\nError: Query required"
+            return "Error: Query is required"
 
         # Resolve model - prefer models with web search capability
         resolved_model, error = await self._resolve_model(
-            model, "text", False, __model__
+            model, "text", False, __model__, __user__
         )
         if error:
-            return f"Venice Web Search\nStatus: 0\nError: {error}"
+            return f"Error: {error}"
 
         if __event_emitter__:
             await __event_emitter__(
@@ -743,16 +799,15 @@ class Tools:
 
             choices = result.get("choices", [])
             if not choices:
-                return f"Venice Web Search ({resolved_model})\nStatus: 200\nError: No response"
+                return f"Error: No response from model {resolved_model}"
 
             assistant_message = choices[0].get("message", {})
             content = assistant_message.get("content", "")
 
             lines = [
-                f"Venice Web Search ({resolved_model})",
+                f"**Venice Web Search** ({resolved_model})",
-                "Status: 200",
                 "",
-                f"Response:\n{content}",
+                f"**Response:**\n{content}",
             ]
 
             # Include citations
@@ -760,20 +815,22 @@ class Tools:
             citations = venice_params.get("web_search_citations", [])
             if citations:
                 lines.append("")
-                lines.append(f"Sources ({len(citations)}):")
+                lines.append(f"**Sources** ({len(citations)}):")
                 for cite in citations[:10]:
                     title = cite.get("title", "Link")
                     url = cite.get("url", "")
-                    lines.append(f"  - {title}")
+                    lines.append(f"- {title}")
                     lines.append(f"  {url}")
 
             return self._truncate("\n".join(lines))
 
         except httpx.HTTPStatusError as e:
+            error_msg = self._format_error(e, f"web search with {resolved_model}")
             if __event_emitter__:
                 await __event_emitter__({"type": "status", "data": {"done": True}})
-            return f"Venice Web Search ({resolved_model})\nStatus: {e.response.status_code}\nError: {e.response.text[:200]}"
+            return f"Error: {error_msg}"
         except Exception as e:
+            error_msg = self._format_error(e, "web search request")
             if __event_emitter__:
                 await __event_emitter__({"type": "status", "data": {"done": True}})
-            return f"Venice Web Search ({resolved_model})\nStatus: 0\nError: {type(e).__name__}: {e}"
+            return f"Error: {error_msg}"