code-puppy 0.0.77__tar.gz → 0.0.79__tar.gz

{code_puppy-0.0.77 → code_puppy-0.0.79}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: code-puppy
-Version: 0.0.77
+Version: 0.0.79
 Summary: Code generation agent
 Author: Michael Pfaffenberger
 License: MIT
@@ -67,8 +67,9 @@ Code Puppy is an AI-powered code generation agent, designed to understand progra
 
 ## Usage
 ```bash
-export MODEL_NAME=gpt-4.1 # or gemini-2.5-flash-preview-05-20 as an example for Google Gemini models
+export MODEL_NAME=gpt-5 # or gemini-2.5-flash-preview-05-20 as an example for Google Gemini models
 export OPENAI_API_KEY=<your_openai_api_key> # or GEMINI_API_KEY for Google Gemini models
+export CEREBRAS_API_KEY=<your_cerebras_api_key> # for Cerebras models
 export YOLO_MODE=true # to bypass the safety confirmation prompt when running shell commands
 
 # or ...
@@ -106,7 +107,7 @@ export MODELS_JSON_PATH=/path/to/custom/models.json
     }
 }
 ```
-Note that the `OPENAI_API_KEY` env variable must be set when using `custom_openai` endpoints.
+Note that the `OPENAI_API_KEY` or `CEREBRAS_API_KEY` env variable must be set when using `custom_openai` endpoints.
 
 Open an issue if your environment is somehow weirder than mine.
 
@@ -122,6 +123,7 @@ code-puppy "write me a C++ hello world program in /tmp/main.cpp then compile it
 - Python 3.9+
 - OpenAI API key (for GPT models)
 - Gemini API key (for Google's Gemini models)
+- Cerebras API key (for Cerebras models)
 - Anthropic key (for Claude models)
 - Ollama endpoint available
 

{code_puppy-0.0.77 → code_puppy-0.0.79}/README.md

@@ -34,8 +34,9 @@ Code Puppy is an AI-powered code generation agent, designed to understand progra
 
 ## Usage
 ```bash
-export MODEL_NAME=gpt-4.1 # or gemini-2.5-flash-preview-05-20 as an example for Google Gemini models
+export MODEL_NAME=gpt-5 # or gemini-2.5-flash-preview-05-20 as an example for Google Gemini models
 export OPENAI_API_KEY=<your_openai_api_key> # or GEMINI_API_KEY for Google Gemini models
+export CEREBRAS_API_KEY=<your_cerebras_api_key> # for Cerebras models
 export YOLO_MODE=true # to bypass the safety confirmation prompt when running shell commands
 
 # or ...
@@ -73,7 +74,7 @@ export MODELS_JSON_PATH=/path/to/custom/models.json
     }
 }
 ```
-Note that the `OPENAI_API_KEY` env variable must be set when using `custom_openai` endpoints.
+Note that the `OPENAI_API_KEY` or `CEREBRAS_API_KEY` env variable must be set when using `custom_openai` endpoints.
 
 Open an issue if your environment is somehow weirder than mine.
 
@@ -89,6 +90,7 @@ code-puppy "write me a C++ hello world program in /tmp/main.cpp then compile it
 - Python 3.9+
 - OpenAI API key (for GPT models)
 - Gemini API key (for Google's Gemini models)
+- Cerebras API key (for Cerebras models)
 - Anthropic key (for Claude models)
 - Ollama endpoint available
 

{code_puppy-0.0.77 → code_puppy-0.0.79}/code_puppy/agent.py

@@ -8,6 +8,7 @@ from pydantic_ai.mcp import MCPServerSSE
 from code_puppy.agent_prompts import get_system_prompt
 from code_puppy.model_factory import ModelFactory
 from code_puppy.session_memory import SessionMemory
+from code_puppy.state_management import message_history_accumulator
 from code_puppy.tools import register_all_tools
 from code_puppy.tools.common import console
 
@@ -83,11 +84,6 @@ def reload_code_generation_agent():
     global _code_generation_agent, _LAST_MODEL_NAME
     from code_puppy.config import get_model_name
 
-    model_name = get_model_name()
-    console.print(f"[bold cyan]Loading Model: {model_name}")
-    global _code_generation_agent, _LAST_MODEL_NAME
-    from code_puppy.config import get_model_name
-
     model_name = get_model_name()
     console.print(f"[bold cyan]Loading Model: {model_name}[/bold cyan]")
     models_path = (
@@ -105,6 +101,7 @@ def reload_code_generation_agent():
         instructions=instructions,
         output_type=str,
         retries=3,
+        history_processors=[message_history_accumulator]
     )
     register_all_tools(agent)
     _code_generation_agent = agent

code_puppy-0.0.79/code_puppy/command_line/motd.py

@@ -0,0 +1,55 @@
+"""
+MOTD (Message of the Day) feature for code-puppy.
+Stores seen versions in ~/.puppy_cfg/motd.txt.
+"""
+
+import os
+
+MOTD_VERSION = "20250815"
+MOTD_MESSAGE = """
+/¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯\\
+|   🐾  Happy Friday, Aug 15, 2025!                        |
+|                                                            |
+|   Biscuit the code puppy is on full zoomie mode!           |
+|   Major paws-ups:                                          |
+|   1. We now integrate Cerebras gpt-oss-120b!               |
+|      It's a bit underwhelming compared to Qwen3-Coder-480b  |
+|      (obviously), but it's still good for basic fetches.   |
+|   2. We also added support for OpenAI gpt-5!                |
+|      It's so good, it'll make you want to teach it to sit!|
+|                                                            |
+|   • To use one of the Cerebras models just have a         |
+|     CEREBRAS_API_KEY set in the environment variables.   |
+|   • Use ~m to swap models in the middle of your session!  |
+|   • Take stretch breaks – you'll need 'em!                 |
+|   • DRY your code, but keep your pup hydrated.            |
+|   • If you hit a bug, treat yourself for finding it!      |
+|                                                            |
+|   Today: sniff, code, roll over, and let these fancy AI   |
+|   models do the heavy lifting. Fire up a ~motd anytime      |
+|   you need some puppy hype!                                |
+\___________________________________________________________/
+"""
+MOTD_TRACK_FILE = os.path.expanduser("~/.puppy_cfg/motd.txt")
+
+
+def has_seen_motd(version: str) -> bool:
+    if not os.path.exists(MOTD_TRACK_FILE):
+        return False
+    with open(MOTD_TRACK_FILE, "r") as f:
+        seen_versions = {line.strip() for line in f if line.strip()}
+    return version in seen_versions
+
+
+def mark_motd_seen(version: str):
+    os.makedirs(os.path.dirname(MOTD_TRACK_FILE), exist_ok=True)
+    with open(MOTD_TRACK_FILE, "a") as f:
+        f.write(f"{version}\n")
+
+
+def print_motd(console, force: bool = False) -> bool:
+    if force or not has_seen_motd(MOTD_VERSION):
+        console.print(MOTD_MESSAGE)
+        mark_motd_seen(MOTD_VERSION)
+        return True
+    return False

{code_puppy-0.0.77 → code_puppy-0.0.79}/code_puppy/main.py

@@ -9,17 +9,19 @@ from rich.markdown import CodeBlock, Markdown
 from rich.syntax import Syntax
 from rich.text import Text
 
-from code_puppy import __version__
+from code_puppy import __version__, state_management
 from code_puppy.agent import get_code_generation_agent, session_memory
 from code_puppy.command_line.prompt_toolkit_completion import (
     get_input_with_combined_completion,
     get_prompt_with_active_model,
 )
 from code_puppy.config import ensure_config_exists
+from code_puppy.state_management import get_message_history, set_message_history
 
 # Initialize rich console for pretty output
 from code_puppy.tools.common import console
 from code_puppy.version_checker import fetch_latest_version
+from code_puppy.message_history_processor import message_history_processor
 
 # from code_puppy.tools import *  # noqa: F403
 
@@ -130,8 +132,6 @@ async def interactive_mode(history_file_path: str) -> None:
                 "[yellow]Falling back to basic input without tab completion[/yellow]"
             )
 
-    message_history = []
-
     # Set up history file in home directory
     history_file_path_prompt = os.path.expanduser("~/.code_puppy_history.txt")
     history_dir = os.path.dirname(history_file_path_prompt)
@@ -172,7 +172,7 @@ async def interactive_mode(history_file_path: str) -> None:
 
         # Check for clear command (supports both `clear` and `~clear`)
         if task.strip().lower() in ("clear", "~clear"):
-            message_history = []
+            state_management._message_history = []
             console.print("[bold yellow]Conversation history cleared![/bold yellow]")
             console.print(
                 "[dim]The agent will not remember previous interactions.[/dim]\n"
@@ -192,71 +192,56 @@ async def interactive_mode(history_file_path: str) -> None:
 
             try:
                 prettier_code_blocks()
-
-                console.log(f"Asking: {task}...", style="cyan")
-
-                # Store agent's full response
-                agent_response = None
-
-                agent = get_code_generation_agent()
-                async with agent.run_mcp_servers():
-                    result = await agent.run(task, message_history=message_history)
-                # Get the structured response
-                agent_response = result.output
-                console.print(agent_response)
-                # Log to session memory
-
-                # Update message history but apply filters & limits
-                new_msgs = result.new_messages()
-                # 1. Drop any system/config messages (e.g., "agent loaded with model")
-                filtered = [
-                    m
-                    for m in new_msgs
-                    if not (isinstance(m, dict) and m.get("role") == "system")
-                ]
-                # 2. Append to existing history and keep only the most recent set by config
-                from code_puppy.config import get_message_history_limit
-
-                message_history.extend(filtered)
-
-                # --- BEGIN GROUP-AWARE TRUNCATION LOGIC ---
-                limit = get_message_history_limit()
-                if len(message_history) > limit:
-
-                    def group_by_tool_call_id(msgs):
-                        grouped = {}
-                        no_group = []
-                        for m in msgs:
-                            # Find all tool_call_id in message parts
-                            tool_call_ids = set()
-                            for part in getattr(m, "parts", []):
-                                if hasattr(part, "tool_call_id") and part.tool_call_id:
-                                    tool_call_ids.add(part.tool_call_id)
-                            if tool_call_ids:
-                                for tcid in tool_call_ids:
-                                    grouped.setdefault(tcid, []).append(m)
-                            else:
-                                no_group.append(m)
-                        return grouped, no_group
-
-                    grouped, no_group = group_by_tool_call_id(message_history)
-                    # Flatten into groups or singletons
-                    grouped_msgs = list(grouped.values()) + [[m] for m in no_group]
-                    # Flattened history (latest groups/singletons last, trunc to N messages total),
-                    # but always keep complete tool_call_id groups together
-                    truncated = []
-                    count = 0
-                    for group in reversed(grouped_msgs):
-                        if count + len(group) > limit:
-                            break
-                        truncated[:0] = group  # insert at front
-                        count += len(group)
-                    message_history = truncated
-                # --- END GROUP-AWARE TRUNCATION LOGIC ---
+                local_cancelled = False
+                async def run_agent_task():
+                    try:
+                        agent = get_code_generation_agent()
+                        async with agent.run_mcp_servers():
+                            return await agent.run(
+                                task,
+                                message_history=get_message_history()
+                            )
+                    except Exception as e:
+                        console.log("Task failed", e)
+
+                agent_task = asyncio.create_task(run_agent_task())
+
+                import signal
+
+                original_handler = None
+
+                def keyboard_interrupt_handler(sig, frame):
+                    nonlocal local_cancelled
+                    if not agent_task.done():
+                        set_message_history(
+                            message_history_processor(
+                                get_message_history()
+                            )
+                        )
+                        agent_task.cancel()
+                        local_cancelled = True
+
+                try:
+                    original_handler = signal.getsignal(signal.SIGINT)
+                    signal.signal(signal.SIGINT, keyboard_interrupt_handler)
+                    result = await agent_task
+                except asyncio.CancelledError:
+                    pass
+                finally:
+                    if original_handler:
+                        signal.signal(signal.SIGINT, original_handler)
+
+                if local_cancelled:
+                    console.print("Task canceled by user")
+                else:
+                    agent_response = result.output
+                    console.print(agent_response)
+                    filtered = message_history_processor(get_message_history())
+                    set_message_history(filtered)
 
                 # Show context status
                 console.print(
-                    f"[dim]Context: {len(message_history)} messages in history[/dim]\n"
+                    f"[dim]Context: {len(get_message_history())} messages in history[/dim]\n"
                 )
 
             except Exception:

code_puppy-0.0.79/code_puppy/message_history_processor.py

@@ -0,0 +1,78 @@
+import queue
+from typing import List
+
+from pydantic_ai.messages import ModelMessage, ToolCallPart, ToolReturnPart
+
+from code_puppy.config import get_message_history_limit
+from code_puppy.tools.common import console
+
+
+def message_history_processor(messages: List[ModelMessage]) -> List[ModelMessage]:
+    """
+    Truncate message history to manage token usage while preserving context.
+
+    This implementation:
+    - Uses the configurable message_history_limit from puppy.cfg (defaults to 40)
+    - Preserves system messages at the beginning
+    - Maintains tool call/response pairs together
+    - Follows PydanticAI best practices for message ordering
+
+    Args:
+        messages: List of ModelMessage objects from conversation history
+
+    Returns:
+        Truncated list of ModelMessage objects
+    """
+    if not messages:
+        return messages
+
+    # Get the configurable limit from puppy.cfg
+    max_messages = get_message_history_limit()
+    # If we have max_messages or fewer, no truncation needed
+    if len(messages) <= max_messages:
+        return messages
+
+    console.print(
+        f"Truncating message history to manage token usage: {max_messages}"
+    )
+    result = []
+    result.append(messages[0])  # this is the system prompt
+    remaining_messages_to_fill = max_messages - 1
+    stack = queue.LifoQueue()
+    count = 0
+    tool_call_parts = set()
+    tool_return_parts = set()
+    for message in reversed(messages):
+        stack.put(message)
+        count += 1
+        if count >= remaining_messages_to_fill:
+            break
+
+    while not stack.empty():
+        item = stack.get()
+        for part in item.parts:
+            if hasattr(part, "tool_call_id") and part.tool_call_id:
+                if isinstance(part, ToolCallPart):
+                    tool_call_parts.add(part.tool_call_id)
+                if isinstance(part, ToolReturnPart):
+                    tool_return_parts.add(part.tool_call_id)
+
+        result.append(item)
+
+    missmatched_tool_call_ids = (tool_call_parts.union(tool_return_parts)) - (
+        tool_call_parts.intersection(tool_return_parts)
+    )
+    # trust...
+    final_result = result
+    if missmatched_tool_call_ids:
+        final_result = []
+        for msg in result:
+            is_missmatched = False
+            for part in msg.parts:
+                if hasattr(part, "tool_call_id"):
+                    if part.tool_call_id in missmatched_tool_call_ids:
+                        is_missmatched = True
+            if is_missmatched:
+                continue
+            final_result.append(msg)
+    return final_result

{code_puppy-0.0.77 → code_puppy-0.0.79}/code_puppy/models.json

@@ -11,6 +11,10 @@
     "type": "openai",
     "name": "gpt-4.1-mini"
   },
+  "gpt-5": {
+    "type": "openai",
+    "name": "gpt-5"
+  },
   "gpt-4.1-nano": {
     "type": "openai",
     "name": "gpt-4.1-nano"
@@ -88,6 +92,14 @@
       "api_key": "$CEREBRAS_API_KEY"
     }
   },
+  "Cerebras-gpt-oss-120b": {
+    "type": "custom_openai",
+    "name": "gpt-oss-120b",
+    "custom_endpoint": {
+      "url": "https://api.cerebras.ai/v1",
+      "api_key": "$CEREBRAS_API_KEY"
+    }
+  },
   "Cerebras-Qwen-3-32b": {
     "type": "custom_openai",
     "name": "qwen-3-32b",

code_puppy-0.0.79/code_puppy/state_management.py

@@ -0,0 +1,42 @@
+from typing import Any, List
+
+from code_puppy.tools.common import console
+
+_message_history: List[Any] = []
+
+def get_message_history() -> List[Any]:
+    return _message_history
+
+def set_message_history(history: List[Any]) -> None:
+    global _message_history
+    _message_history = history
+
+def clear_message_history() -> None:
+    global _message_history
+    _message_history = []
+
+def append_to_message_history(message: Any) -> None:
+    _message_history.append(message)
+
+def extend_message_history(history: List[Any]) -> None:
+    _message_history.extend(history)
+
+
+def hash_message(message):
+    hashable_entities = []
+    for part in message.parts:
+        if hasattr(part, "timestamp"):
+            hashable_entities.append(part.timestamp.isoformat())
+        elif hasattr(part, "tool_call_id"):
+            hashable_entities.append(part.tool_call_id)
+        else:
+            hashable_entities.append(part.content)
+    return hash(",".join(hashable_entities))
+
+
+def message_history_accumulator(messages: List[Any]):
+    message_history_hashes = set([hash_message(m) for m in _message_history])
+    for msg in messages:
+        if hash_message(msg) not in message_history_hashes:
+            _message_history.append(msg)
+    return messages

{code_puppy-0.0.77 → code_puppy-0.0.79}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "code-puppy"
-version = "0.0.77"
+version = "0.0.79"
 description = "Code generation agent"
 readme = "README.md"
 requires-python = ">=3.10"

code_puppy-0.0.77/code_puppy/command_line/motd.py

@@ -1,49 +0,0 @@
-"""
-MOTD (Message of the Day) feature for code-puppy.
-Stores seen versions in ~/.puppy_cfg/motd.txt.
-"""
-
-import os
-
-MOTD_VERSION = "20250802"
-MOTD_MESSAGE = """
-/¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯\\
-|   🐾  Happy Sat-urday, Aug 2, 2025!                       |
-|                                                           |
-|   Biscuit the code puppy is on full zoomie mode!          |
-|   Major paws-up: We now integrate Cerebras Qwen3 Coder    |
-|   480b! YES, that’s 480 billion parameters of tail-wagging|
-|   code speed. It’s so fast, even my fetch can’t keep up!  |
-|                                                           |
-|   • Take stretch breaks – you’ll need ‘em!                |
-|   • DRY your code, but keep your pup hydrated.            |
-|   • If you hit a bug, treat yourself for finding it!      |
-|                                                           |
-|   Today: sniff, code, roll over, and let Cerebras Qwen3   |
-|   Coder 480b do the heavy lifting. Fire up a ~motd anytime|
-|   you need some puppy hype!                               |
-\___________________________________________________________/
-"""
-MOTD_TRACK_FILE = os.path.expanduser("~/.puppy_cfg/motd.txt")
-
-
-def has_seen_motd(version: str) -> bool:
-    if not os.path.exists(MOTD_TRACK_FILE):
-        return False
-    with open(MOTD_TRACK_FILE, "r") as f:
-        seen_versions = {line.strip() for line in f if line.strip()}
-    return version in seen_versions
-
-
-def mark_motd_seen(version: str):
-    os.makedirs(os.path.dirname(MOTD_TRACK_FILE), exist_ok=True)
-    with open(MOTD_TRACK_FILE, "a") as f:
-        f.write(f"{version}\n")
-
-
-def print_motd(console, force: bool = False) -> bool:
-    if force or not has_seen_motd(MOTD_VERSION):
-        console.print(MOTD_MESSAGE)
-        mark_motd_seen(MOTD_VERSION)
-        return True
-    return False