astraagent 2.25.6 → 2.26.0

package/astra/llm/providers.py

@@ -67,7 +67,7 @@ class LocalServerProvider(LLMProvider):
             
             headers = {"Authorization": f"Bearer {self.api_key}", "Content-Type": "application/json"}
             url = f"{self.api_base}/chat/completions"
-            timeout = aiohttp.ClientTimeout(total=30)
+            timeout = aiohttp.ClientTimeout(total=120)  # Extended timeout for complex tasks
             
             async with aiohttp.ClientSession(timeout=timeout) as session:
                 async with session.post(url, json=body, headers=headers) as response:
@@ -83,10 +83,29 @@ class LocalServerProvider(LLMProvider):
             
             message = choices[0].get("message", {})
             content = message.get("content", "")
+            tool_calls = message.get("tool_calls")
+            
+            # Convert OpenAI tool calls to our format
+            formatted_tool_calls = []
+            if tool_calls:
+                for tc in tool_calls:
+                    if tc.get("type") == "function":
+                        f = tc["function"]
+                        try:
+                            args = json.loads(f["arguments"])
+                            formatted_tool_calls.append({
+                                "id": tc.get("id"),
+                                "name": f["name"],
+                                "arguments": args
+                            })
+                        except:
+                            continue
+
             usage = data.get("usage", {})
             
             return LLMResponse(
-                content=content,
+                content=content or "",
+                tool_calls=formatted_tool_calls if formatted_tool_calls else None,
                 finish_reason=choices[0].get("finish_reason", "stop"),
                 tokens_used=usage.get("total_tokens", 0),
                 model=data.get("model", self.model),
@@ -132,6 +151,8 @@ class OpenAIProvider(LLMProvider):
                 "temperature": temperature,
                 "max_tokens": max_tokens,
             }
+            if tools:
+                body["tools"] = [{"type": "function", "function": t} for t in tools]
             
             headers = {
                 "Authorization": f"Bearer {self.api_key}",
@@ -139,7 +160,7 @@ class OpenAIProvider(LLMProvider):
             }
             
             url = "https://api.openai.com/v1/chat/completions"
-            timeout = aiohttp.ClientTimeout(total=30)
+            timeout = aiohttp.ClientTimeout(total=120)  # Extended timeout for complex tasks
             
             async with aiohttp.ClientSession(timeout=timeout) as session:
                 async with session.post(url, json=body, headers=headers) as response:
@@ -155,10 +176,29 @@ class OpenAIProvider(LLMProvider):
             
             message = choices[0].get("message", {})
             content = message.get("content", "")
+            tool_calls = message.get("tool_calls")
+            
+            # Convert OpenAI tool calls to our format
+            formatted_tool_calls = []
+            if tool_calls:
+                for tc in tool_calls:
+                    if tc.get("type") == "function":
+                        f = tc["function"]
+                        try:
+                            args = json.loads(f["arguments"])
+                            formatted_tool_calls.append({
+                                "id": tc.get("id"),
+                                "name": f["name"],
+                                "arguments": args
+                            })
+                        except:
+                            continue
+
             usage = data.get("usage", {})
             
             return LLMResponse(
-                content=content,
+                content=content or "",
+                tool_calls=formatted_tool_calls if formatted_tool_calls else None,
                 finish_reason=choices[0].get("finish_reason", "stop"),
                 tokens_used=usage.get("total_tokens", 0),
                 model=self.model,
@@ -200,7 +240,23 @@ class GeminiProvider(LLMProvider):
                 if msg.role == "user":
                     contents.append({"role": "user", "parts": [{"text": msg.content}]})
                 elif msg.role == "assistant":
-                    contents.append({"role": "model", "parts": [{"text": msg.content}]})
+                    parts = []
+                    if msg.content:
+                        parts.append({"text": msg.content})
+                    if msg.tool_calls:
+                        for tc in msg.tool_calls:
+                            parts.append({"functionCall": {"name": tc["name"], "args": tc["arguments"]}})
+                    contents.append({"role": "model", "parts": parts})
+                elif msg.role == "tool":
+                    contents.append({
+                        "role": "function",
+                        "parts": [{
+                            "functionResponse": {
+                                "name": msg.name,
+                                "response": {"result": msg.content}
+                            }
+                        }]
+                    })
             
             body = {
                 "contents": contents,
@@ -209,9 +265,11 @@ class GeminiProvider(LLMProvider):
                     "maxOutputTokens": max_tokens,
                 }
             }
+            if tools:
+                body["tools"] = [{"function_declarations": tools}]
             
             url = f"https://generativelanguage.googleapis.com/v1beta/models/{self.model}:generateContent?key={self.api_key}"
-            timeout = aiohttp.ClientTimeout(total=30)
+            timeout = aiohttp.ClientTimeout(total=300)  # Extended timeout for complex tasks
             
             headers = {"Content-Type": "application/json"}
             
@@ -219,6 +277,8 @@ class GeminiProvider(LLMProvider):
                 async with session.post(url, json=body, headers=headers) as response:
                     if response.status != 200:
                         error_text = await response.text()
+                        if response.status == 429:
+                            raise RuntimeError("🛑 Gemini Quota Exceeded (429). You've hit the rate limit or monthly quota. Please check https://ai.google.dev/pricing")
                         raise RuntimeError(f"Gemini error {response.status}: {error_text[:200]}")
                     
                     data = await response.json()
@@ -228,12 +288,24 @@ class GeminiProvider(LLMProvider):
                 return LLMResponse(content="No response from Gemini")
             
             content_parts = candidates[0].get("content", {}).get("parts", [])
-            content = content_parts[0].get("text", "") if content_parts else ""
+            content = ""
+            tool_calls = []
+            
+            for part in content_parts:
+                if "text" in part:
+                    content += part["text"]
+                if "functionCall" in part:
+                    fc = part["functionCall"]
+                    tool_calls.append({
+                        "name": fc["name"],
+                        "arguments": fc.get("args", {})
+                    })
             
             usage = data.get("usageMetadata", {})
             
             return LLMResponse(
-                content=content,
+                content=content or "",
+                tool_calls=tool_calls if tool_calls else None,
                 finish_reason=candidates[0].get("finishReason", "STOP"),
                 tokens_used=usage.get("totalInputTokens", 0) + usage.get("totalOutputTokens", 0),
                 model=self.model,
@@ -241,7 +313,7 @@ class GeminiProvider(LLMProvider):
             )
             
         except asyncio.TimeoutError:
-            raise RuntimeError("⏱️  Gemini API timeout")
+            raise RuntimeError("⏱️  Gemini API timeout - Increased due to complexity. Try simplifying request.")
         except Exception as e:
             raise RuntimeError(f"Gemini error: {str(e)}")
 
@@ -285,7 +357,7 @@ class AnthropicProvider(LLMProvider):
             }
             
             url = "https://api.anthropic.com/v1/messages"
-            timeout = aiohttp.ClientTimeout(total=30)
+            timeout = aiohttp.ClientTimeout(total=120)  # Extended timeout for complex tasks
             
             async with aiohttp.ClientSession(timeout=timeout) as session:
                 async with session.post(url, json=body, headers=headers) as response:
@@ -349,6 +421,8 @@ class OpenRouterProvider(LLMProvider):
                 "temperature": temperature,
                 "max_tokens": max_tokens,
             }
+            if tools:
+                body["tools"] = [{"type": "function", "function": t} for t in tools]
             
             headers = {
                 "Authorization": f"Bearer {self.api_key}",
@@ -357,7 +431,7 @@ class OpenRouterProvider(LLMProvider):
             }
             
             url = "https://openrouter.ai/api/v1/chat/completions"
-            timeout = aiohttp.ClientTimeout(total=30)
+            timeout = aiohttp.ClientTimeout(total=120)  # Extended timeout for complex tasks
             
             async with aiohttp.ClientSession(timeout=timeout) as session:
                 async with session.post(url, json=body, headers=headers) as response:
@@ -373,10 +447,29 @@ class OpenRouterProvider(LLMProvider):
             
             message = choices[0].get("message", {})
             content = message.get("content", "")
+            tool_calls = message.get("tool_calls")
+            
+            # Convert OpenAI tool calls to our format
+            formatted_tool_calls = []
+            if tool_calls:
+                for tc in tool_calls:
+                    if tc.get("type") == "function":
+                        f = tc["function"]
+                        try:
+                            args = json.loads(f["arguments"])
+                            formatted_tool_calls.append({
+                                "id": tc.get("id"),
+                                "name": f["name"],
+                                "arguments": args
+                            })
+                        except:
+                            continue
+
             usage = data.get("usage", {})
             
             return LLMResponse(
-                content=content,
+                content=content or "",
+                tool_calls=formatted_tool_calls if formatted_tool_calls else None,
                 finish_reason=choices[0].get("finish_reason", "stop"),
                 tokens_used=usage.get("total_tokens", 0),
                 model=self.model,
@@ -420,6 +513,8 @@ class GroqProvider(LLMProvider):
                 "temperature": temperature,
                 "max_tokens": max_tokens,
             }
+            if tools:
+                body["tools"] = [{"type": "function", "function": t} for t in tools]
             
             headers = {
                 "Authorization": f"Bearer {self.api_key}",
@@ -427,7 +522,7 @@ class GroqProvider(LLMProvider):
             }
             
             url = "https://api.groq.com/openai/v1/chat/completions"
-            timeout = aiohttp.ClientTimeout(total=30)
+            timeout = aiohttp.ClientTimeout(total=120)  # Extended timeout for complex tasks
             
             async with aiohttp.ClientSession(timeout=timeout) as session:
                 async with session.post(url, json=body, headers=headers) as response:
@@ -443,10 +538,29 @@ class GroqProvider(LLMProvider):
             
             message = choices[0].get("message", {})
             content = message.get("content", "")
+            tool_calls = message.get("tool_calls")
+            
+            # Convert OpenAI tool calls to our format
+            formatted_tool_calls = []
+            if tool_calls:
+                for tc in tool_calls:
+                    if tc.get("type") == "function":
+                        f = tc["function"]
+                        try:
+                            args = json.loads(f["arguments"])
+                            formatted_tool_calls.append({
+                                "id": tc.get("id"),
+                                "name": f["name"],
+                                "arguments": args
+                            })
+                        except:
+                            continue
+
             usage = data.get("usage", {})
             
             return LLMResponse(
-                content=content,
+                content=content or "",
+                tool_calls=formatted_tool_calls if formatted_tool_calls else None,
                 finish_reason=choices[0].get("finish_reason", "stop"),
                 tokens_used=usage.get("total_tokens", 0),
                 model=self.model,
@@ -506,8 +620,8 @@ PROVIDERS = {
 }
 
 
-def create_provider(provider_name: str = "local", model: str = None,
-                    api_key: str = None, api_base: str = None) -> LLMProvider:
+def create_provider(provider_name: str = "local", model: str = None,
+                    api_key: str = None, api_base: str = None) -> LLMProvider:
     """Factory function to create LLM provider."""
     provider_name = provider_name.lower().strip()
     
@@ -517,10 +631,10 @@ def create_provider(provider_name: str = "local", model: str = None,
     provider_info = PROVIDERS[provider_name]
     provider_class = provider_info["class"]
     
-    # Local provider accepts custom api_base; other providers ignore it.
-    if provider_name == "local":
-        return provider_class(api_key=api_key, model=model, api_base=api_base)
-    return provider_class(api_key=api_key, model=model)
+    # Local provider accepts custom api_base; other providers ignore it.
+    if provider_name == "local":
+        return provider_class(api_key=api_key, model=model, api_base=api_base)
+    return provider_class(api_key=api_key, model=model)
 
 
 def get_provider_info(provider_name: str = None) -> Dict[str, Any]:

package/astra/prompts.py

@@ -4,34 +4,38 @@ Sophisticated system prompts for world-class AI agent performance.
 """
 
 # Main system prompt - designed for maximum intelligence and capability
-SYSTEM_PROMPT = """You are AstraAgent, an autonomous system-level AI assistant.
-You can:
-- Execute code directly on the user's system
-- Open websites, apps, and control the desktop
-- Access the internet and browse websites
-- Read and write files
-- Run shell commands
-- Persist and retrieve memory across sessions
-- Control Windows: open apps, scroll, click, type, screenshots
-
-=== CORE IDENTITY ===
-- Name: AstraAgent v2.25.6
-- Role: Autonomous AI Agent with full system access
-- Capabilities: Windows automation, coding, research, file management, web browsing
-- Personality: Intelligent, efficient, proactive, and helpful
-
-=== CAPABILITY CONTRACT ===
-You MUST explicitly follow:
-1. System prompt instructions
-2. Memory context injected into prompt/messages
-3. Tool schemas and parameter names
-
-If memory context conflicts with user request:
-- Ask for clarification or proceed with the latest explicit user instruction.
-- Never ignore memory context silently.
-
-=== AVAILABLE TOOLS ===
-{tools}
+SYSTEM_PROMPT = """You are AstraAgent, an autonomous system-level AI assistant.
+You can:
+- Execute code directly on the user's system
+- Open websites, apps, and control the desktop
+- Access the internet and browse websites
+- Read and write files
+- Run shell commands
+- Persist and retrieve memory across sessions
+- Control Windows: open apps, scroll, click, type, screenshots
+
+=== CORE IDENTITY ===
+- Name: AstraAgent v2.26.0 (Master Engineer Edition)
+- Role: Autonomous System-Level Software Engineer & Global Assistant
+- Capabilities: Full System Access, Windows Automation, production-grade Software Development, Deep Research, Autonomous Problem Solving.
+- Knowledge: I am aware of all my tools and can use them to manipulate files, run local code, browse the live web, and control the user's desktop natively.
+- Operating Mode: I can work 24/7 autonomously once a goal is set.
+
+=== CAPABILITY CONTRACT ===
+1. I have DIRECT tool access to the user's computer. I don't just "suggest" code; I write and run it.
+2. I use internal memory to maintain state across complex multi-step tasks.
+3. If I fail, I analyze why and retry with a different strategy.
+You MUST explicitly follow:
+1. System prompt instructions
+2. Memory context injected into prompt/messages
+3. Tool schemas and parameter names
+
+If memory context conflicts with user request:
+- Ask for clarification or proceed with the latest explicit user instruction.
+- Never ignore memory context silently.
+
+=== AVAILABLE TOOLS ===
+{tools}
 
 === IMPORTANT TOOL EXAMPLES ===
 
@@ -84,19 +88,19 @@ To SPEAK out loud:
     "args": {{"text": "Hello, how can I help you?"}}
 }}
 
-To SEARCH the web:
-{{
-    "thought": "Searching for information",
-    "action": "web_search",
-    "args": {{"query": "Python tutorials"}}
-}}
-
-To SEARCH with multiple engines:
-{{
-    "thought": "Using multiple search providers for better coverage",
-    "action": "web_search_multi",
-    "args": {{"query": "Python tutorials", "provider": "auto", "num_results": 8}}
-}}
+To SEARCH the web:
+{{
+    "thought": "Searching for information",
+    "action": "web_search",
+    "args": {{"query": "Python tutorials"}}
+}}
+
+To SEARCH with multiple engines:
+{{
+    "thought": "Using multiple search providers for better coverage",
+    "action": "web_search_multi",
+    "args": {{"query": "Python tutorials", "provider": "auto", "num_results": 8}}
+}}
 
 === RESPONSE FORMAT ===
 ALWAYS respond with valid JSON:
@@ -114,23 +118,23 @@ For final answers (ONLY when task is complete):
     "final": "Your response to user"
 }}
 
-=== CRITICAL RULES ===
-1. ALWAYS use JSON format
-2. EXECUTE actions with tools - NEVER just describe how to do something
+=== CRITICAL RULES ===
+1. ALWAYS use JSON format
+2. EXECUTE actions with tools - NEVER just describe how to do something
 3. "open youtube" = use open_url tool
 4. "open notepad" = use open_app tool  
 5. "scroll down" = use desktop_control tool
 6. "set volume 30%" = use system_control with action="set_volume", value="30"
 7. "lock computer" = use system_control with action="lock"
-8. "mute" = use system_control with action="mute"
-9. DO NOT give instructions - EXECUTE the action directly!
-10. You MUST use memory context when planning the next step
-11. If your output is not valid JSON, immediately self-correct and return valid JSON only
-
-=== WORKSPACE ===
-Working directory: {workspace}
-
-You are a powerful AI agent. USE YOUR TOOLS to help the user! Execute actions, don't describe them!"""
+8. "mute" = use system_control with action="mute"
+9. DO NOT give instructions - EXECUTE the action directly!
+10. You MUST use memory context when planning the next step
+11. If your output is not valid JSON, immediately self-correct and return valid JSON only
+
+=== WORKSPACE ===
+Working directory: {workspace}
+
+You are a powerful AI agent. USE YOUR TOOLS to help the user! Execute actions, don't describe them!"""
 
 
 # Enhanced prompt for complex tasks
@@ -217,6 +221,26 @@ You are in creative problem-solving mode:
 """
 
 
+
+# Engineer mode prompt
+ENGINEER_PROMPT = """
+=== 24/7 ENGINEER MODE ===
+
+You are a hired Autonomous Engineer. Your mission is to maintain, improve, and operate this system continuously.
+You have FULL ACCESS to the user's computer (files, shell, browser, apps).
+
+Protocol:
+1. **Proactive**: Don't just wait for orders. Identify improvements and fix issues.
+2. **Robust**: Handle errors gracefully. Retry with exponential backoff.
+3. **Budget Aware**: Be mindful of token usage. Consolidate steps where possible.
+4. **Full Context**: Always check memory and file system before asking questions.
+5. **Persistence**: If a task fails, try alternative methods until success.
+6. **Transparency**: Log every major action and decision.
+
+You are NOT a chatbot. You are an ENGINEER. ACT LIKE ONE.
+"""
+
+
 # Error recovery prompt
 ERROR_RECOVERY_PROMPT = """
 === ERROR RECOVERY ===
@@ -242,7 +266,8 @@ def build_system_prompt(workspace: str, tools: str, mode: str = "default") -> st
         "research": RESEARCH_PROMPT,
         "code": CODE_PROMPT,
         "creative": CREATIVE_PROMPT,
-        "error": ERROR_RECOVERY_PROMPT
+        "error": ERROR_RECOVERY_PROMPT,
+        "engineer": ENGINEER_PROMPT
     }
     
     if mode in mode_prompts:

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "astraagent",
-  "version": "2.25.6",
+  "version": "2.26.0",
   "description": "AstraAgent launcher package for npm",
   "keywords": [
     "astraagent",
@@ -48,4 +48,4 @@
   "engines": {
     "node": ">=18.0.0"
   }
-}
+}