code-lm 0.3.0__tar.gz → 0.3.1__tar.gz

{code_lm-0.3.0/src/code_lm.egg-info → code_lm-0.3.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: code-lm
-Version: 0.3.0
+Version: 0.3.1
 Summary: An AI coding assistant using various LLM models.
 Home-page: https://github.com/Panagiotis897/lm-code
 Author: Panagiotis897
@@ -47,6 +47,12 @@ LM Code is a powerful AI coding assistant for your terminal supporting 17 free m
   - Test running: `pytest` and similar tools.
 - **Customizable Configurations**:
   - Easily set default models and API keys.
+- **Session Persistence**:
+  - Conversation history is saved per project directory.
+  - Resume previous sessions when you restart in the same directory.
+- **Mid-Session Commands**:
+  - `/compact` — Summarize conversation history to stay within context limits.
+  - `/model` — Switch models mid-session without restarting.
 
 ---
 
@@ -134,6 +140,9 @@ During an interactive session:
 
 - **`/exit`**: Exit the session.
 - **`/help`**: Display help information.
+- **`/compact`**: Summarize the conversation history to reduce token usage. Useful when the session gets long and you want to free up context space.
+- **`/model`**: List all available models.
+- **`/model <model_id>`**: Switch to a different model mid-session without losing conversation history.
 
 ---
 
@@ -153,10 +162,16 @@ LM Code is under active development. Contributions, feature requests, and feedba
 
 ### Changelog
 
+#### v0.3.1
+- Added `/compact` command — summarizes conversation history via LLM to stay within context limits.
+- Added `/model` command — list available models or switch models mid-session without restarting.
+- Added session persistence — conversation history is auto-saved per project directory and can be resumed on restart.
+- Updated help text to include new interactive commands.
+
 #### v0.3.0
 - Updated default model to NVIDIA Nemotron 3 Super 120B.
 - Added 17 free models from OpenRouter (previously 6).
-- Fixed `ModuleNotFoundError: No module named 'gemini_cli'` from stale entry point.
+- Fixed `ModuleNotFoundError: No module named 'gemini_cli'` from stale entry point after package rename.
 - Fixed `UnicodeDecodeError` on Windows (cp1253) for all subprocess commands.
 - Fixed API URL (`/chat/completions` was missing) causing HTML response errors.
 - Improved API error handling for empty/invalid responses.
@@ -168,20 +183,18 @@ LM Code is under active development. Contributions, feature requests, and feedba
 #### v0.2.5
 - Added more models to the model list.
 - Fixed crucial bugs from previous versions.
-- Removed Gemini models.
+- Removed legacy Gemini module.
 - Updated models to latest versions.
 
 #### v0.1.0
-- Rebranded from Gemini to LM Code.
+- Rebranded to LM Code.
 - Integrated OpenRouter as the default provider.
 - Added multi-model support.
-- Overhauled CLI commands (`gemini` -> `lmcode`).
+- Overhauled CLI commands.
 
 ---
 
 ## Future Plans
-
-- Pricing with appropriate rate limits.
 - Non-free model support.
 - MCP Server integration.
 - Additional providers.

{code_lm-0.3.0 → code_lm-0.3.1}/README.md

@@ -19,6 +19,12 @@ LM Code is a powerful AI coding assistant for your terminal supporting 17 free m
   - Test running: `pytest` and similar tools.
 - **Customizable Configurations**:
   - Easily set default models and API keys.
+- **Session Persistence**:
+  - Conversation history is saved per project directory.
+  - Resume previous sessions when you restart in the same directory.
+- **Mid-Session Commands**:
+  - `/compact` — Summarize conversation history to stay within context limits.
+  - `/model` — Switch models mid-session without restarting.
 
 ---
 
@@ -106,6 +112,9 @@ During an interactive session:
 
 - **`/exit`**: Exit the session.
 - **`/help`**: Display help information.
+- **`/compact`**: Summarize the conversation history to reduce token usage. Useful when the session gets long and you want to free up context space.
+- **`/model`**: List all available models.
+- **`/model <model_id>`**: Switch to a different model mid-session without losing conversation history.
 
 ---
 
@@ -125,10 +134,16 @@ LM Code is under active development. Contributions, feature requests, and feedba
 
 ### Changelog
 
+#### v0.3.1
+- Added `/compact` command — summarizes conversation history via LLM to stay within context limits.
+- Added `/model` command — list available models or switch models mid-session without restarting.
+- Added session persistence — conversation history is auto-saved per project directory and can be resumed on restart.
+- Updated help text to include new interactive commands.
+
 #### v0.3.0
 - Updated default model to NVIDIA Nemotron 3 Super 120B.
 - Added 17 free models from OpenRouter (previously 6).
-- Fixed `ModuleNotFoundError: No module named 'gemini_cli'` from stale entry point.
+- Fixed `ModuleNotFoundError: No module named 'gemini_cli'` from stale entry point after package rename.
 - Fixed `UnicodeDecodeError` on Windows (cp1253) for all subprocess commands.
 - Fixed API URL (`/chat/completions` was missing) causing HTML response errors.
 - Improved API error handling for empty/invalid responses.
@@ -140,20 +155,18 @@ LM Code is under active development. Contributions, feature requests, and feedba
 #### v0.2.5
 - Added more models to the model list.
 - Fixed crucial bugs from previous versions.
-- Removed Gemini models.
+- Removed legacy Gemini module.
 - Updated models to latest versions.
 
 #### v0.1.0
-- Rebranded from Gemini to LM Code.
+- Rebranded to LM Code.
 - Integrated OpenRouter as the default provider.
 - Added multi-model support.
-- Overhauled CLI commands (`gemini` -> `lmcode`).
+- Overhauled CLI commands.
 
 ---
 
 ## Future Plans
-
-- Pricing with appropriate rate limits.
 - Non-free model support.
 - MCP Server integration.
 - Additional providers.

{code_lm-0.3.0 → code_lm-0.3.1}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "code-lm"
-version = "0.3.0"
+version = "0.3.1"
 authors = [
   { name="Panagiotis897", email="orion256business@gmail.com" }
 ]

{code_lm-0.3.0 → code_lm-0.3.1/src/code_lm.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: code-lm
-Version: 0.3.0
+Version: 0.3.1
 Summary: An AI coding assistant using various LLM models.
 Home-page: https://github.com/Panagiotis897/lm-code
 Author: Panagiotis897
@@ -47,6 +47,12 @@ LM Code is a powerful AI coding assistant for your terminal supporting 17 free m
   - Test running: `pytest` and similar tools.
 - **Customizable Configurations**:
   - Easily set default models and API keys.
+- **Session Persistence**:
+  - Conversation history is saved per project directory.
+  - Resume previous sessions when you restart in the same directory.
+- **Mid-Session Commands**:
+  - `/compact` — Summarize conversation history to stay within context limits.
+  - `/model` — Switch models mid-session without restarting.
 
 ---
 
@@ -134,6 +140,9 @@ During an interactive session:
 
 - **`/exit`**: Exit the session.
 - **`/help`**: Display help information.
+- **`/compact`**: Summarize the conversation history to reduce token usage. Useful when the session gets long and you want to free up context space.
+- **`/model`**: List all available models.
+- **`/model <model_id>`**: Switch to a different model mid-session without losing conversation history.
 
 ---
 
@@ -153,10 +162,16 @@ LM Code is under active development. Contributions, feature requests, and feedba
 
 ### Changelog
 
+#### v0.3.1
+- Added `/compact` command — summarizes conversation history via LLM to stay within context limits.
+- Added `/model` command — list available models or switch models mid-session without restarting.
+- Added session persistence — conversation history is auto-saved per project directory and can be resumed on restart.
+- Updated help text to include new interactive commands.
+
 #### v0.3.0
 - Updated default model to NVIDIA Nemotron 3 Super 120B.
 - Added 17 free models from OpenRouter (previously 6).
-- Fixed `ModuleNotFoundError: No module named 'gemini_cli'` from stale entry point.
+- Fixed `ModuleNotFoundError: No module named 'gemini_cli'` from stale entry point after package rename.
 - Fixed `UnicodeDecodeError` on Windows (cp1253) for all subprocess commands.
 - Fixed API URL (`/chat/completions` was missing) causing HTML response errors.
 - Improved API error handling for empty/invalid responses.
@@ -168,20 +183,18 @@ LM Code is under active development. Contributions, feature requests, and feedba
 #### v0.2.5
 - Added more models to the model list.
 - Fixed crucial bugs from previous versions.
-- Removed Gemini models.
+- Removed legacy Gemini module.
 - Updated models to latest versions.
 
 #### v0.1.0
-- Rebranded from Gemini to LM Code.
+- Rebranded to LM Code.
 - Integrated OpenRouter as the default provider.
 - Added multi-model support.
-- Overhauled CLI commands (`gemini` -> `lmcode`).
+- Overhauled CLI commands.
 
 ---
 
 ## Future Plans
-
-- Pricing with appropriate rate limits.
 - Non-free model support.
 - MCP Server integration.
 - Additional providers.

{code_lm-0.3.0 → code_lm-0.3.1}/src/code_lm.egg-info/SOURCES.txt

@@ -11,9 +11,9 @@ src/code_lm.egg-info/top_level.txt
 src/lm_code/__init__.py
 src/lm_code/config.py
 src/lm_code/main.py
+src/lm_code/session.py
 src/lm_code/utils.py
 src/lm_code/models/__init__.py
-src/lm_code/models/gemini.py
 src/lm_code/models/openrouter.py
 src/lm_code/tools/__init__.py
 src/lm_code/tools/base.py

{code_lm-0.3.0 → code_lm-0.3.1}/src/lm_code/__init__.py

@@ -2,4 +2,4 @@
 LM CLI - A command-line interface for interacting with various LLM models.
 """
 
-__version__ = "0.1.70"
+__version__ = "0.3.1"

{code_lm-0.3.0 → code_lm-0.3.1}/src/lm_code/main.py

@@ -13,11 +13,13 @@ from pathlib import Path
 import yaml
 import logging
 import time
+import questionary
 
 from .models.openrouter import OpenRouterModel, list_available_models
 from .config import Config
 from .utils import count_tokens
 from .tools import AVAILABLE_TOOLS
+from .session import SessionManager
 
 # Setup console and config
 console = Console(
@@ -259,11 +261,51 @@ def start_interactive_session(model_name: str, console: Console):
         )
         return
 
+    # --- Session persistence setup ---
+    session_mgr = SessionManager()
+    project_dir = str(Path.cwd())
+    saved_session = session_mgr.load_session(project_dir)
+
+    if saved_session:
+        saved_model = saved_session.get("model_name", "?")
+        saved_time = time.strftime(
+            "%Y-%m-%d %H:%M",
+            time.localtime(saved_session.get("timestamp", 0)),
+        )
+        msg_count = len(saved_session.get("chat_history", []))
+        console.print(
+            f"\n[yellow]Found saved session[/yellow] (model: {saved_model}, "
+            f"saved: {saved_time}, {msg_count} messages)"
+        )
+        try:
+            resume = questionary.confirm(
+                "Resume previous session?",
+                default=True,
+                auto_enter=False,
+            ).ask()
+        except (KeyboardInterrupt, EOFError):
+            resume = None
+
+        if resume is None:
+            console.print("\n[yellow]Exiting.[/yellow]")
+            return
+    else:
+        resume = False
+    # ---
+
     try:
         console.print(f"\nInitializing model [bold]{model_name}[/bold]...")
         # Pass the console object to OpenRouterModel constructor
         model = OpenRouterModel(api_key=api_key, console=console, model_name=model_name)
-        console.print("[green]Model initialized successfully.[/green]\n")
+
+        # Restore session history if resuming
+        if resume and saved_session:
+            model.chat_history = saved_session.get("chat_history", model.chat_history)
+            console.print(
+                f"[green]Session restored ({len(model.chat_history)} messages).[/green]\n"
+            )
+        else:
+            console.print("[green]Model initialized successfully.[/green]\n")
 
     except Exception as e:
         console.print(
@@ -277,6 +319,7 @@ def start_interactive_session(model_name: str, console: Console):
 
     # --- Session Start Message ---
     console.print("Type '/help' for commands, '/exit' or Ctrl+C to quit.")
+    console.print(f"[dim]Current model: {model.current_model_name}[/dim]")
 
     while True:
         try:
@@ -287,6 +330,9 @@ def start_interactive_session(model_name: str, console: Console):
             elif user_input.lower() == "/help":
                 show_help()
                 continue
+            elif user_input.lower().startswith("/model"):
+                _handle_model_command(user_input, model, console, SUPPORTED_MODELS)
+                continue
 
             # Display initial "thinking" status - generate handles intermediate ones
             response_text = model.generate(user_input)
@@ -299,8 +345,17 @@ def start_interactive_session(model_name: str, console: Console):
                 log.warning("generate() returned None unexpectedly.")
                 continue
 
-            console.print("[bold green]Assistant:[/bold green]")
-            console.print(Markdown(response_text), highlight=True)
+            # Show compact result without markdown rendering
+            if user_input.lower().strip() == "/compact":
+                console.print(f"[cyan]{response_text}[/cyan]")
+            else:
+                console.print("[bold green]Assistant:[/bold green]")
+                console.print(Markdown(response_text), highlight=True)
+
+            # Auto-save session after each turn
+            session_mgr.save_session(
+                project_dir, model.chat_history, model.current_model_name
+            )
 
         except KeyboardInterrupt:
             console.print("\n[yellow]Session interrupted. Exiting.[/yellow]")
@@ -311,6 +366,43 @@ def start_interactive_session(model_name: str, console: Console):
             )
             log.error("Error during interactive loop", exc_info=True)
 
+    # Save session on exit
+    session_mgr.save_session(project_dir, model.chat_history, model.current_model_name)
+    console.print("[dim]Session saved.[/dim]")
+
+
+def _handle_model_command(
+    user_input: str, model, console: Console, supported_models: list
+):
+    """Handle /model command to switch models mid-session."""
+    parts = user_input.strip().split(None, 1)
+    if len(parts) < 2:
+        # No model specified — show available models
+        console.print("[cyan]Available models:[/cyan]")
+        for m in supported_models:
+            current = (
+                " [bold green](current)[/bold green]"
+                if m["id"] == model.current_model_name
+                else ""
+            )
+            console.print(f"  [bold]{m['id']}[/bold]{current} — {m['description']}")
+        console.print("\nUsage: /model <model_id>")
+        return
+
+    target = parts[1].strip()
+
+    # Validate the model is in supported list
+    valid_ids = {m["id"] for m in supported_models}
+    if target not in valid_ids:
+        console.print(f"[red]Unknown model:[/red] {target}")
+        console.print("Use [bold]/model[/bold] (no argument) to list available models.")
+        return
+
+    old_model = model.current_model_name
+    model.switch_model(target)
+    console.print(f"[green]Switched model:[/green] {old_model} → [bold]{target}[/bold]")
+    console.print(f"[dim]Current model: {model.current_model_name}[/dim]")
+
 
 def show_help():
     """Show help information for interactive mode."""
@@ -329,6 +421,9 @@ def show_help():
  [cyan]Interactive Commands:[/cyan]
   /exit
   /help
+  /compact          Summarize conversation history to save tokens
+  /model            List available models
+  /model <name>     Switch to a different model mid-session
 
  [cyan]CLI Commands:[/cyan]
   lmcode setup KEY

{code_lm-0.3.0 → code_lm-0.3.1}/src/lm_code/models/openrouter.py

@@ -188,6 +188,99 @@ class OpenRouterModel:
             )
         return tools
 
+    # --- Model switching ---
+    def switch_model(self, model_name: str) -> None:
+        """Switch to a different model without resetting conversation history."""
+        self.current_model_name = model_name
+        log.info(f"Switched model to: {model_name}")
+
+    # --- Compact history via summarization ---
+    def compact_history(self) -> str:
+        """Summarize the conversation history to reduce token usage."""
+        if len(self.chat_history) <= 2:
+            return "Nothing to compact."
+
+        self.console.print("[yellow]Compacting conversation history...[/yellow]")
+
+        # Extract user-facing conversation (skip system message at index 0)
+        conversation_text = ""
+        for msg in self.chat_history[1:]:
+            role = msg.get("role", "?")
+            content = msg.get("content", "")
+            if role == "user":
+                # Strip the orientation context prefix for cleaner summary
+                lines = content.split("\n")
+                user_lines = []
+                capturing = False
+                for line in lines:
+                    if line.startswith("User request:"):
+                        capturing = True
+                    if capturing:
+                        user_lines.append(line)
+                if user_lines:
+                    conversation_text += f"User: {' '.join(user_lines)}\n"
+                else:
+                    conversation_text += f"User: {content[:500]}\n"
+            elif role == "assistant":
+                conversation_text += f"Assistant: {content[:1000]}\n"
+            elif role == "tool":
+                name = msg.get("name", "tool")
+                conversation_text += f"Tool({name}): {content[:500]}\n"
+
+        if not conversation_text.strip():
+            return "Nothing to compact."
+
+        summarization_prompt = (
+            "You are a summarizer. Summarize the following conversation between a user and an AI coding assistant. "
+            "Preserve all key details: what was discussed, what files were viewed/edited, "
+            "what decisions were made, and what the current state of work is. "
+            "Be concise but complete. This summary will replace the conversation history.\n\n"
+            f"Conversation:\n{conversation_text}\n\n"
+            "Provide a concise summary:"
+        )
+
+        try:
+            payload = {
+                "model": self.current_model_name,
+                "messages": [
+                    {
+                        "role": "system",
+                        "content": "You are a helpful summarizer. Return only the summary, no preamble.",
+                    },
+                    {"role": "user", "content": summarization_prompt},
+                ],
+                "temperature": 0.3,
+                "max_tokens": 1000,
+            }
+
+            with self.console.status("[yellow]Generating summary...", spinner="dots"):
+                response = requests.post(
+                    self.base_url, headers=self.headers, json=payload
+                )
+                response.raise_for_status()
+                data = response.json()
+
+            summary = data["choices"][0]["message"]["content"]
+            original_count = len(self.chat_history)
+
+            # Keep system message, replace everything else with summary
+            self.chat_history = [
+                self.chat_history[0],  # system message
+                {
+                    "role": "assistant",
+                    "content": f"[Conversation summary]\n{summary}",
+                },
+            ]
+
+            log.info(
+                f"Compacted history from {original_count} to {len(self.chat_history)} messages"
+            )
+            return f"Compacted {original_count} messages down to a summary."
+
+        except Exception as e:
+            log.error(f"Compaction failed: {e}", exc_info=True)
+            return f"Compaction failed: {e}"
+
     # --- Native Function Calling Agent Loop ---
     def generate(self, prompt: str) -> str | None:
         logging.info(
@@ -196,10 +289,11 @@ class OpenRouterModel:
         original_user_prompt = prompt
         if prompt.startswith("/"):
             command = prompt.split()[0].lower()
-            # Handle commands like /compact here eventually
             if command in ["/exit", "/help"]:
                 logging.info(f"Handled command: {command}")
-                return None  # Or return specific help text
+                return None
+            if command == "/compact":
+                return self.compact_history()
 
         # === Step 1: Mandatory Orientation ===
         orientation_context = ""

code_lm-0.3.1/src/lm_code/session.py

@@ -0,0 +1,97 @@
+"""
+Session persistence for LM Code.
+Saves and loads conversation history per project directory.
+"""
+
+import hashlib
+import json
+import logging
+import os
+import time
+from pathlib import Path
+from typing import Optional, Dict, List, Any
+
+log = logging.getLogger(__name__)
+
+SESSIONS_DIR = Path.home() / ".config" / "lm-code" / "sessions"
+
+
+def _project_hash(project_dir: str) -> str:
+    """Create a stable hash from the absolute project directory path."""
+    abs_path = str(Path(project_dir).resolve())
+    return hashlib.sha256(abs_path.encode()).hexdigest()[:16]
+
+
+class SessionManager:
+    """Manages per-project session persistence."""
+
+    def __init__(self):
+        SESSIONS_DIR.mkdir(parents=True, exist_ok=True)
+
+    def _session_path(self, project_dir: str) -> Path:
+        h = _project_hash(project_dir)
+        return SESSIONS_DIR / f"{h}.json"
+
+    def save_session(
+        self,
+        project_dir: str,
+        chat_history: List[Dict[str, Any]],
+        model_name: str,
+    ) -> None:
+        """Save chat history and metadata for a project directory."""
+        data = {
+            "project_dir": str(Path(project_dir).resolve()),
+            "model_name": model_name,
+            "timestamp": time.time(),
+            "chat_history": chat_history,
+        }
+        path = self._session_path(project_dir)
+        try:
+            with open(path, "w", encoding="utf-8") as f:
+                json.dump(data, f, ensure_ascii=False, indent=2)
+            log.debug(f"Session saved to {path}")
+        except Exception as e:
+            log.error(f"Failed to save session: {e}")
+
+    def load_session(self, project_dir: str) -> Optional[Dict[str, Any]]:
+        """Load a saved session for a project directory, or None."""
+        path = self._session_path(project_dir)
+        if not path.exists():
+            return None
+        try:
+            with open(path, "r", encoding="utf-8") as f:
+                data = json.load(f)
+            log.info(f"Session loaded from {path}")
+            return data
+        except Exception as e:
+            log.error(f"Failed to load session: {e}")
+            return None
+
+    def delete_session(self, project_dir: str) -> bool:
+        """Delete a saved session. Returns True if one was deleted."""
+        path = self._session_path(project_dir)
+        if path.exists():
+            path.unlink()
+            log.info(f"Session deleted: {path}")
+            return True
+        return False
+
+    def list_sessions(self) -> List[Dict[str, Any]]:
+        """List all saved sessions with metadata."""
+        sessions = []
+        for p in SESSIONS_DIR.glob("*.json"):
+            try:
+                with open(p, "r", encoding="utf-8") as f:
+                    data = json.load(f)
+                sessions.append(
+                    {
+                        "project_dir": data.get("project_dir", "?"),
+                        "model_name": data.get("model_name", "?"),
+                        "timestamp": data.get("timestamp", 0),
+                        "message_count": len(data.get("chat_history", [])),
+                    }
+                )
+            except Exception:
+                continue
+        sessions.sort(key=lambda s: s["timestamp"], reverse=True)
+        return sessions

{code_lm-0.3.0 → code_lm-0.3.1}/src/lm_code/utils.py

@@ -9,9 +9,7 @@ import json
 def count_tokens(text):
     """
     Count the number of tokens in a text string.
-
-    This is a rough estimate for Gemini 2.5 Pro, using GPT-4 tokenizer as a proxy.
-    For production, you'd want to use model-specific token counting.
+    Uses GPT-4 tokenizer as a proxy.
     """
     try:
         encoding = tiktoken.encoding_for_model("gpt-4")

code_lm-0.3.0/src/lm_code/models/gemini.py

@@ -1,43 +0,0 @@
-"""
-Gemini model integration for the CLI tool.
-"""
-
-import logging
-import time
-from rich.console import Console
-from rich.panel import Panel
-import questionary
-
-from ..utils import count_tokens
-from ..tools import get_tool, AVAILABLE_TOOLS
-
-# Setup logging (basic config, consider moving to main.py)
-logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - [%(filename)s:%(lineno)d] - %(message)s')
-log = logging.getLogger(__name__)
-
-MAX_AGENT_ITERATIONS = 10
-CONTEXT_TRUNCATION_THRESHOLD_TOKENS = 800000  # Example token limit
-
-
-class GeminiModel:
-    """Interface for Gemini models using native function calling agentic loop."""
-
-    def __init__(self, console: Console):
-        """Initialize the Gemini model interface."""
-        self.console = console
-
-        # --- Tool Definition ---
-        self.function_declarations = None  # Tools have been removed
-        # ---
-
-        # --- System Prompt (Native Functions & Planning) ---
-        self.system_instruction = "Initialize system prompt."
-        # ---
-
-        # --- Initialize Persistent History ---
-        self.chat_history = [
-            {'role': 'user', 'parts': [self.system_instruction]},
-            {'role': 'model', 'parts': ["Okay, I'm ready. Provide the directory context and your request."]}
-        ]
-        log.info("Initialized persistent chat history.")
-        # ---