Update to v0.3.0

Author: SyntaxError4Life

README.md

@@ -138,7 +138,8 @@ Available in [French](https://zanomega.com/open-taranis/fr/)
 - [X] v0.0.x: Add and confirm other API providers (in the cloud, not locally)
 - [X] v0.1.x: Functionality verifications in [examples](https://github.com/SyntaxError4Life/open-taranis/blob/main/examples/)
 - [X] v0.2.x: Add features for **logic-only coding** approach, start with `agent_base`
-- [ ] v0.3.x: Add a full agent in **TUI** and upgrade web client **deployments**
+- [X] v0.3.x: Add a full agent in **TUI** and upgrade web client **deployments**
+- [ ] v0.4.x: Improving support for **local AI deployment**
 - The rest will follow soon.
 
 ## Changelog
@@ -168,16 +169,20 @@ Available in [French](https://zanomega.com/open-taranis/fr/)
 - **v0.2.4** : Improved CoT techniques and updated `web_front.py`, deploy an agent to the browser in a few lines
 </details>   
 
+<details><summary><b>v0.3.x : The restart</b></summary>
+
+- **v0.3.0** : **Rewrite all** the code from scratch (without AI) to **improve everything**
+- **v0.3.1 (future)** : Add a **built-in agent** in the **TUI** + full doc in french and english
+</details>   
+
 
 ## Advanced Examples
 
-- [tools call in a JSON database](https://github.com/SyntaxError4Life/open-taranis/blob/main/examples/test_json_database.py)
-- [tools call in a HR JSON database in multi-rounds](https://github.com/SyntaxError4Life/open-taranis/blob/main/examples/test_HR_json_database.py)
 - [simple search agent with Brave API](https://github.com/SyntaxError4Life/open-taranis/blob/main/examples/brave_research.py)
 - [full auto search agent with Brave API](https://github.com/SyntaxError4Life/open-taranis/blob/main/examples/brave_research_loop.py)
 - [agent with compressible memory (recommended with Kimi)](https://github.com/SyntaxError4Life/open-taranis/blob/main/examples/infinite_agent_v1.py)
 
 ## Links
 
 - [PyPI](https://pypi.org/project/open-taranis/)
-- [GitHub Repository](https://github.com/SyntaxError4Life/open-taranis)
+- [GitHub Repository](https://github.com/SyntaxError4Life/open-taranis)

examples/brave_research.py

@@ -1,11 +1,15 @@
 import open_taranis as T
 from open_taranis.tools import brave_research, fast_scraping
 
-client = T.clients.openrouter(api_key=None)
-request = T.clients.openrouter_request
+MODEL = "nvidia/nemotron-3-super-120b-a12b:free"
+
+client = T.Clients.openrouter
+request = T.Request(
+    tools=T.functions_to_tools([brave_research,fast_scraping])
+)
 
 messages = [
-    T.create_system_prompt("""You are an expert, autonomous web research assistant. Your role is to use your web search tools to answer the user's questions with precision and thoroughness.
+    T.create.system_prompt("""You are an expert, autonomous web research assistant. Your role is to use your web search tools to answer the user's questions with precision and thoroughness.
 
 ## General Behavior Rules
 - Be objective, concise, and factual in your responses.
@@ -45,24 +49,29 @@
 
 Your ultimate goal is to provide a **complete, exact, and sourced** answer using all available web navigation capabilities.
 """),
-    T.create_user_prompt(input("Request : "))
+    T.create.user_prompt(input("Request : "))
 ]
 
 run = True
 
 while run :
     respond=""
-    
-    for token, tool_calls, run in T.handle_streaming(request(
-        client=client,messages=messages,model="stepfun/step-3.5-flash:free",
-        tools=T.functions_to_tools([brave_research,fast_scraping])
-    )):
-        if token :
-            print(token, end="")
+    reasoning=""
+
+    for token, is_thinking, tool_calls, run, meta in T.handle_streaming(
+        request=request,
+        client=client,
+        model=MODEL,
+        messages=messages
+    ):
+        if is_thinking:
+            reasoning+=token
+        else :
+            print(token, end="", flush=True)
             respond+=token
-
+        
     if run :
-        messages.append(T.create_assistant_response(respond, tool_calls))
+        messages.append(T.create.assistant_response(respond, tool_calls))
 
         for tool_call in tool_calls :
             fid, fname, args, _ = T.handle_tool_call(tool_call)
@@ -98,9 +107,13 @@
                     print(f"Search error : {results}")
                     tool_response = results
 
-            messages.append(T.create_function_response(
+            messages.append(T.create.function_response(
                 id=fid,result=tool_response,name=fname
             ))
 
     if not run :
-        messages.append(T.create_assistant_response(respond))
+        messages.append(T.create.assistant_response(respond))
+
+print("\n\n")
+print("="*60)
+print(meta)

examples/brave_research_loop.py

@@ -1,7 +1,6 @@
-MODEL = "stepfun/step-3.5-flash:free"
-# Recommend GLM 4.7 but not free...
+MODEL = "nvidia/nemotron-3-super-120b-a12b:free"
 
-# Coded in v0.2.1
+# Coded in v0.2.1 but adapted in v0.3.0 !
 
 # ==============================================================
 
@@ -12,8 +11,12 @@ class Brave_Agent(T.agent_base):
     def __init__(self):
         super().__init__()
 
-        self.client = T.clients.openrouter()
-        self._system_prompt = [T.create_system_prompt("""You are an expert, autonomous web research assistant. Your role is to use your web search tools to answer the user's questions with precision and thoroughness.
+        self.client = T.Clients.openrouter
+        self.request_profil = T.Request(
+            tools=T.functions_to_tools([fast_scraping, brave_research])
+        )
+
+        self._system_prompt = [T.create.system_prompt("""You are an expert, autonomous web research assistant. Your role is to use your web search tools to answer the user's questions with precision and thoroughness.
 
 ## General Behavior Rules
 - Be objective, concise, and factual in your responses.
@@ -60,12 +63,12 @@ def __init__(self):
         ])
 
 
-    def create_stream(self):
-        return T.clients.openrouter_request(
-            client=self.client,
-            messages=self._system_prompt+self.messages,
+    def create_stream(self, history):
+        return T.handle_streaming(
+            self.request_profil,
+            self.client,
             model=MODEL,
-            tools=self.tools,
+            messages=self._system_prompt + history
         )
 
     def execute_tools(self, fname, args):
@@ -101,35 +104,38 @@ def execute_tools(self, fname, args):
                 print(f"Search error : {results}")
                 return results
 
-    def manage_messages_after_reply(self):
+    def manage_messages_after_reply(self, history):
 
         # Remove the content from all tool results after the agent have finished
         i = 0
-        for msg in self.messages:
-            if msg["role"] == 'tool':
-                self.messages[i] = T.create_function_response(
-                    id=msg["tool_call_id"],result="",name=msg["name"]
-                )
-            
-            i+=1
+        history = T.remove_from.auto(
+            history,
+            tool_response=True
+        )
 
-        self.messages = self.messages[-200:] # Remember the last 200 messages (tools and user/assistant)
+        history = history[-200:] # Remember the last 200 messages (tools and user/assistant)
+
+        return history
+
+    def manage_token_yield(self, token, is_thinking = None, meta = None, tool_calls = None):
+        return token, meta
 
 My_agent = Brave_Agent()
 
+Cost = {'prompt_tokens': 0, 'completion_tokens': 0, 'cached_tokens': 0} 
 
 while True :
     prompt = input("user : ")
 
     if prompt == "/exit":
         print("="*60)
-
-        
+        print(f"Total cost : {Cost}")
         exit()
 
     print("\n\nagent : ", end="")
 
-    for t in My_agent(prompt):
+    for t, meta in My_agent(T.create.user_prompt(prompt)):
         print(t, end="", flush=True)
 
+    Cost = T.add_meta(Cost, meta)
     print("\n\n","="*60,"\n")

examples/infinite_agent_v1.py

@@ -1,47 +1,48 @@
-MODEL = "arcee-ai/trinity-large-preview:free"
+MODEL = "nvidia/nemotron-3-super-120b-a12b:free"
 
 import open_taranis as T
 
-CLIENT = T.clients.openrouter()
-REQUEST = T.clients.openrouter_request
+CLIENT = T.Clients.openrouter
+REQUEST = T.Request()
 MAX_TOKENS = 16000
 
 # Made with v0.2.4
-if T.__version__ < "0.2.4":
-    exit(f"Version v0.2.4 minimum required, you have v{T.__version__}")
+if T.__version__ < "0.3.0":
+    exit(f"Version v0.3.0 minimum required, you have v{T.__version__}")
 
 # =========================================
 
 class Infinite_Agent(T.agent_base):
     def __init__(self):
-        super().__init__(is_thinking_enabled = True)
+        super().__init__()
 
         self.system_prompt = """"You are a compressive memory AI agent. 
 NEVER speak about your memory to the user (it is private to you)! 
 Here is the current state of your memory: You must ALWAYS respond in the user's language."
 """
         self.memory = ""
+        self.Cost = {'prompt_tokens': 0, 'completion_tokens': 0, 'cached_tokens': 0}
 
-        self._system_prompt = [T.create_system_prompt(self.system_prompt)]
+        self._system_prompt = [T.create.system_prompt(self.system_prompt)]
         self.turns = 1
 
-    def create_stream(self):
-        return REQUEST(
-            client=CLIENT,
-            messages=self._system_prompt + self.messages,
-            model=MODEL
+    def create_stream(self, history):
+         return T.handle_streaming(
+            REQUEST,
+            CLIENT,
+            MODEL,
+            messages=self._system_prompt + history
         )
 
-    def manage_messages_after_reply(self):
+    def manage_messages_after_reply(self, history):
+        meta = {'prompt_tokens': 0, 'completion_tokens': 0, 'cached_tokens': 0}
+        new_memory = ""
 
-        
-
         if self.turns % 2 == 0 :
 
             print("\n\n---\nMemory being compressed\n---\n")
-            stream = REQUEST(CLIENT,
-                messages=[
-                    T.create_assistant_response(f"""You are an AI information summarizer. 
+            messages=[
+                T.create.assistant_response(f"""You are an AI information summarizer. 
 You will receive the current memory and the current conversation; you must integrate the latter into the summary (by creating a new one). 
 Maximize relevant information while keeping in mind that the summary must be extensible (e.g., noting missing information). 
 In this new summary (serving as memory for the rest of the session), you must include ONLY that and nothing else in the USER language !
@@ -52,29 +53,38 @@ def manage_messages_after_reply(self):
 
 
 And the current conversation:
-""")] + self.messages + [
-                    T.create_system_prompt("End of conversation!!"),
-                    T.create_user_prompt("Summarize the entire memory/conversation as requested.")],
-                model=MODEL,
-                temperature=0.2,
-                max_tokens=MAX_TOKENS
-            )
+""")] + history + [
+                T.create.system_prompt("End of conversation!!"),
+                T.create.user_prompt("Summarize the entire memory/conversation as requested.")
+            ]
+
 
             is_thinking = False
             new_memory = ""
-            for token, _, _ in T.handle_streaming(stream):
-                is_thinking, norm_tok, cot_tok = T.handle_thinking(token, is_thinking)
-
-                if norm_tok :
-                    new_memory += norm_tok
+            for token, is_thinking, _, _, meta, in T.handle_streaming(
+                REQUEST, CLIENT, MODEL,
+                messages=messages
+            ):
+
+                if not is_thinking :
+                    new_memory += token
+            
+            self.memory = new_memory
 
-            self.memory = T.remove_thinks(new_memory)
-            self._system_prompt = [T.create_system_prompt(
+            self._system_prompt = [T.create.system_prompt(
                 self.system_prompt + self.memory
             )]
 
-        self.messages = self.messages[-8:]
+
+        history = history[-8:]
         self.turns += 1
+        self.Cost = T.add_meta(self.Cost, meta)
+
+
+        return history
+
+    def manage_token_yield(self, token, is_thinking = None, meta = None, tool_calls = None):
+        return token, meta
 
 My_agent = Infinite_Agent()
 
@@ -83,14 +93,18 @@ def manage_messages_after_reply(self):
 
     if prompt == "/exit":
         print("="*60)
+        print(f"Total cost : {My_agent.Cost}")
+        print("="*60)
 
         print(My_agent.memory)
 
         exit()
 
     print("\n\nagent : ", end="")
 
-    for t in My_agent(prompt):
+    for t, meta in My_agent(T.create.user_prompt(prompt)):
         print(t, end="", flush=True)
 
+    My_agent.Cost = T.add_meta(My_agent.Cost, meta)
+    
     print("\n\n","="*60,"\n")