gitai-local 1.3.0__tar.gz → 1.3.2__tar.gz

{gitai_local-1.3.0 → gitai_local-1.3.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: gitai-local
-Version: 1.3.0
+Version: 1.3.2
 Summary: AI-powered conventional commit message generator running 100% locally
 License-Expression: MIT
 Project-URL: Homepage, https://github.com/PabloSalinasDev/GitAI-CLI
@@ -64,7 +64,7 @@ The benefit of implementing this CLI is twofold:
 3. After staging changes, run `gitai`. It feeds the diff (or initial commit context) into the model and proposes a commit message.
 4. At the end of your session, run `gitai out` to stop the daemon and free RAM.
 
-### 📖 Interactive Help & Documentation Access
+### Interactive Help & Documentation Access
 
 GitAI CLI features a native, robust command-line validation and help module. If you ever need a quick reminder of the available architecture commands, explicit routing links, or want to report a technical issue, you can invoke the user manual directly from your shell.
 
@@ -206,9 +206,9 @@ Based on extensive stress-testing on standard consumer hardware (e.g., Intel i7
 
 | Hardware Profile | Context Size | Estimated Time * | Commit Accuracy |
 |------------------|--------------|------------------|-----------------|
-| **Modern Desktop / Laptop**<br>*(Ryzen 5/7, Core i5/i7 11th+ Gen, DDR5)* | Small to Large Diffs | **~8 – 20 seconds** | ~98% |
-| **Older / Standard Hardware**<br>*(Legacy Intel i7, DDR4 @ 2400MHz)* | Small to Medium Diffs | **~20 – 40 seconds** | ~98% |
-| **Stress Test / Massive Changes**<br>*(Legacy Hardware + Dense Diffs)* | Large Multi-file Diffs | **~40 – 70 seconds** | ~98% |
+| **Modern Desktop / Laptop**<br>*(Ryzen 5/7, Core i5/i7 11th+ Gen, DDR5)* | Small to Large Diffs | **~8 – 20 seconds** | ~90% |
+| **Older / Standard Hardware**<br>*(Legacy Intel i7, DDR4 @ 2400MHz)* | Small to Medium Diffs | **~20 – 40 seconds** | ~90% |
+| **Stress Test / Massive Changes**<br>*(Legacy Hardware + Dense Diffs)* | Large Multi-file Diffs | **~40 – 70 seconds** | ~85% |
 
 > **Note on Performance:** Times are heavily bound to RAM clock speed and single-core efficiency. A dense, multi-file diff analyzed on legacy hardware might occasionally touch the upper boundary of the stress test zone. However, GitAI's built-in traffic manager ensures payloads remain strictly bounded to keep local execution controlled and predictable.
 
@@ -218,7 +218,7 @@ Based on extensive stress-testing on standard consumer hardware (e.g., Intel i7
 
 Why are these times so consistent? GitAI does not just dump raw data into the LLM. It includes a custom pre-processing pipeline that strips out compiler noise, binary files, white spaces, and structural brackets before sending the payload. 
 
-* **Linear vs. Deductive Processing:** Testing proved that sending a complete, clean code block up to 2500 characters is significantly faster than truncating it too early. By providing the model with full, clean context, the LLM processes the data linearly instead of wasting CPU cycles trying to "guess" missing code structures. This balance cuts down processing overhead by up to 30 seconds on older machines while boosting commit accuracy to a staggering **98%**.
+* **Linear vs. Deductive Processing:** Testing proved that sending a complete, clean code block up to 2500 characters is significantly faster than truncating it too early. By providing the model with full, clean context, the LLM processes the data linearly instead of wasting CPU cycles trying to "guess" missing code structures. This balance cuts down processing overhead by up to 40 seconds on older machines while boosting commit accuracy to a staggering **85%**.
 
 ## Dependencies
 
@@ -231,6 +231,33 @@ Why are these times so consistent? GitAI does not just dump raw data into the LL
 
 ---
 
+## Privacy & Security
+
+GitAI is designed with a strict privacy-first architecture:
+
+* **100% Local Execution:** All prompt processing and commit analysis run entirely on your local CPU/GPU using `llama.cpp` bindings.
+* **Zero Telemetry:** The CLI collects no analytics, logs, or usage metrics.
+* **Air-Gapped Friendly:** Once the GGUF model is downloaded, no internet connection is required to generate commits or run commands.
+* **No Code Exposure:** Your repository diffs and code never leave your machine or hit external APIs.
+
+---
+
+## Uninstallation
+
+To completely remove GitAI and delete the downloaded local model weights (~4.7 GB):
+
+```bash
+# 1. Delete the local model cache
+Remove-Item -Recurse -Force "$env:LOCALAPPDATA\gitai"   # Windows
+
+# 2. Uninstall the package
+pipx uninstall gitai-local    # If installed via pipx (README instructions)
+# or
+pip uninstall gitai-local     # If installed via standard pip (PyPI default)
+```
+
+---
+
 ## License
 
 This project is licensed under the MIT License. See the [LICENSE](LICENSE) file for details.

{gitai_local-1.3.0 → gitai_local-1.3.2}/README.md

@@ -43,7 +43,7 @@ The benefit of implementing this CLI is twofold:
 3. After staging changes, run `gitai`. It feeds the diff (or initial commit context) into the model and proposes a commit message.
 4. At the end of your session, run `gitai out` to stop the daemon and free RAM.
 
-### 📖 Interactive Help & Documentation Access
+### Interactive Help & Documentation Access
 
 GitAI CLI features a native, robust command-line validation and help module. If you ever need a quick reminder of the available architecture commands, explicit routing links, or want to report a technical issue, you can invoke the user manual directly from your shell.
 
@@ -185,9 +185,9 @@ Based on extensive stress-testing on standard consumer hardware (e.g., Intel i7
 
 | Hardware Profile | Context Size | Estimated Time * | Commit Accuracy |
 |------------------|--------------|------------------|-----------------|
-| **Modern Desktop / Laptop**<br>*(Ryzen 5/7, Core i5/i7 11th+ Gen, DDR5)* | Small to Large Diffs | **~8 – 20 seconds** | ~98% |
-| **Older / Standard Hardware**<br>*(Legacy Intel i7, DDR4 @ 2400MHz)* | Small to Medium Diffs | **~20 – 40 seconds** | ~98% |
-| **Stress Test / Massive Changes**<br>*(Legacy Hardware + Dense Diffs)* | Large Multi-file Diffs | **~40 – 70 seconds** | ~98% |
+| **Modern Desktop / Laptop**<br>*(Ryzen 5/7, Core i5/i7 11th+ Gen, DDR5)* | Small to Large Diffs | **~8 – 20 seconds** | ~90% |
+| **Older / Standard Hardware**<br>*(Legacy Intel i7, DDR4 @ 2400MHz)* | Small to Medium Diffs | **~20 – 40 seconds** | ~90% |
+| **Stress Test / Massive Changes**<br>*(Legacy Hardware + Dense Diffs)* | Large Multi-file Diffs | **~40 – 70 seconds** | ~85% |
 
 > **Note on Performance:** Times are heavily bound to RAM clock speed and single-core efficiency. A dense, multi-file diff analyzed on legacy hardware might occasionally touch the upper boundary of the stress test zone. However, GitAI's built-in traffic manager ensures payloads remain strictly bounded to keep local execution controlled and predictable.
 
@@ -197,7 +197,7 @@ Based on extensive stress-testing on standard consumer hardware (e.g., Intel i7
 
 Why are these times so consistent? GitAI does not just dump raw data into the LLM. It includes a custom pre-processing pipeline that strips out compiler noise, binary files, white spaces, and structural brackets before sending the payload. 
 
-* **Linear vs. Deductive Processing:** Testing proved that sending a complete, clean code block up to 2500 characters is significantly faster than truncating it too early. By providing the model with full, clean context, the LLM processes the data linearly instead of wasting CPU cycles trying to "guess" missing code structures. This balance cuts down processing overhead by up to 30 seconds on older machines while boosting commit accuracy to a staggering **98%**.
+* **Linear vs. Deductive Processing:** Testing proved that sending a complete, clean code block up to 2500 characters is significantly faster than truncating it too early. By providing the model with full, clean context, the LLM processes the data linearly instead of wasting CPU cycles trying to "guess" missing code structures. This balance cuts down processing overhead by up to 40 seconds on older machines while boosting commit accuracy to a staggering **85%**.
 
 ## Dependencies
 
@@ -210,6 +210,33 @@ Why are these times so consistent? GitAI does not just dump raw data into the LL
 
 ---
 
+## Privacy & Security
+
+GitAI is designed with a strict privacy-first architecture:
+
+* **100% Local Execution:** All prompt processing and commit analysis run entirely on your local CPU/GPU using `llama.cpp` bindings.
+* **Zero Telemetry:** The CLI collects no analytics, logs, or usage metrics.
+* **Air-Gapped Friendly:** Once the GGUF model is downloaded, no internet connection is required to generate commits or run commands.
+* **No Code Exposure:** Your repository diffs and code never leave your machine or hit external APIs.
+
+---
+
+## Uninstallation
+
+To completely remove GitAI and delete the downloaded local model weights (~4.7 GB):
+
+```bash
+# 1. Delete the local model cache
+Remove-Item -Recurse -Force "$env:LOCALAPPDATA\gitai"   # Windows
+
+# 2. Uninstall the package
+pipx uninstall gitai-local    # If installed via pipx (README instructions)
+# or
+pip uninstall gitai-local     # If installed via standard pip (PyPI default)
+```
+
+---
+
 ## License
 
 This project is licensed under the MIT License. See the [LICENSE](LICENSE) file for details.

{gitai_local-1.3.0 → gitai_local-1.3.2}/gitai/config.py

@@ -33,7 +33,7 @@ def ensure_model():
     print(Fore.CYAN + "│   GitAI - First run: downloading AI model   │")
     print(Fore.CYAN + "│      AI Model Setup - one time only         │")
     print(Fore.CYAN + "│                                             │")
-    print(Fore.CYAN + "│ Developed by PyBloSoft © 2026 - Ver. 1.3.0  │")
+    print(Fore.CYAN + "│ Developed by PyBloSoft © 2026 - Ver. 1.3.2  │")
     print(Fore.CYAN + "└─────────────────────────────────────────────┘\n")
 
     MODEL_DIR.mkdir(parents=True, exist_ok=True)
@@ -50,25 +50,27 @@ def ensure_model():
 
     def print_progress(pct, d_gb, t_gb):
         bar_len = 30
-        filled  = int(bar_len * pct)
-        bar     = Fore.YELLOW + "█" * filled + Style.DIM + "░" * (bar_len - filled)
-        print(Fore.YELLOW + f"\r[{bar}" + Style.RESET_ALL + Fore.YELLOW + f"] {pct*100:.1f}%  {d_gb:.2f}/{t_gb:.2f} GB", end="", flush=True)
+        filled = int(bar_len * pct)
+        bar = "█" * filled + "░" * (bar_len - filled)
+        
+        line = f"\r\033[K[{bar}] {pct*100:.1f}%  {d_gb:.2f}/{t_gb:.2f} GB"
+        sys.stdout.write(Fore.YELLOW + line + Style.RESET_ALL)
+        sys.stdout.flush()
 
-    f = None
     try:
         with open(tmp_path, "wb") as f:
             with httpx.stream("GET", MODEL_URL, follow_redirects=True, timeout=None) as r:
                 r.raise_for_status()
-                total      = int(r.headers.get("content-length", 0))
+                total = int(r.headers.get("content-length", 0))
                 downloaded = 0
 
                 for chunk in r.iter_bytes(chunk_size=1024 * 256):
                     f.write(chunk)
                     downloaded += len(chunk)
                     if total:
-                        pct  = downloaded / total
+                        pct = downloaded / total
                         d_gb = downloaded / 1_073_741_824
-                        t_gb = total      / 1_073_741_824
+                        t_gb = total / 1_073_741_824
                         print_progress(pct, d_gb, t_gb)
 
         sys.stdout.write("\033[?25h")
@@ -78,13 +80,6 @@ def ensure_model():
         print(Fore.GREEN + "\n\n Model downloaded successfully.\n")
 
     except KeyboardInterrupt:
-
-        if f and not f.closed:
-            try:
-                f.close()
-            except Exception:
-                pass
-        
         sys.stdout.write("\033[?25h")
         sys.stdout.flush()
         
@@ -98,13 +93,6 @@ def ensure_model():
         os._exit(1)
 
     except Exception as e:
-
-        if f and not f.closed:
-            try:
-                f.close()
-            except Exception:
-                pass
-                
         sys.stdout.write("\033[?25h")
         sys.stdout.flush()
 

{gitai_local-1.3.0 → gitai_local-1.3.2}/gitai/llm_client.py

@@ -10,7 +10,6 @@ import threading
 from colorama import init, Fore, Style
 
 from gitai.config import MODEL_PATH, PORT
-import __main__
 
 init(autoreset=True)
 
@@ -112,14 +111,16 @@ def clean_git_diff(raw_diff, max_estimated=2500, debug=True):
                     percent = 95
 
                 filled_length = int(bar_length * percent // 100)
-                bar = Fore.YELLOW + '█' * filled_length + Style.DIM + '░' * (bar_length - filled_length)
+                bar = '█' * filled_length + '░' * (bar_length - filled_length)
 
-                sys.stdout.write(Fore.YELLOW + f'\r Crunching diff: [{bar}' + Style.RESET_ALL + Fore.YELLOW + f'] {percent:.0f}%')
+                line = f"\r\033[K Crunching diff: [{bar}] {percent:.0f}%"
+                sys.stdout.write(Fore.YELLOW + line + Style.RESET_ALL)
                 sys.stdout.flush()
                 time.sleep(0.2)
 
             bar_final = '█' * bar_length
-            sys.stdout.write(Fore.YELLOW + f'\r Crunching diff: [{bar_final}' + Fore.YELLOW + '] 100%\n\n')
+            final_line = f"\r\033[K Crunching diff: [{bar_final}] 100%\n\n"
+            sys.stdout.write(Fore.YELLOW + final_line + Style.RESET_ALL)
             sys.stdout.flush()
             
         finally:

{gitai_local-1.3.0 → gitai_local-1.3.2}/gitai/main.py

@@ -7,6 +7,18 @@ from gitai.llm_client import start_daemon, stop_daemon, generate_commit_message
 
 init(autoreset=True)
 
+def is_git_repo():
+    result = subprocess.run(
+        ["git", "rev-parse", "--is-inside-work-tree"],
+        capture_output=True,
+        text=True,
+        check=False,
+        encoding="utf-8",
+        errors="replace",
+        env=get_secure_env()
+    )
+    return result.returncode == 0
+
 def is_initial_commit():
     result = subprocess.run(
         ["git", "rev-parse", "HEAD"],
@@ -82,21 +94,27 @@ def run_commit(message):
 
 def print_help():
     """Print the user manual integrated directly into the terminal with web access."""
-    print(Fore.CYAN + "═"*74)
-    print(Fore.CYAN + "         GitAI CLI - Offline Privacy-First Commit Generator")
-    print(Fore.CYAN + "             Developed by PyBloSoft © 2026 - Ver. 1.3.0")
-    print(Fore.CYAN + "═"*74)
+
+    print(Fore.CYAN + "═"*86)
+    print(Fore.CYAN + "                 GitAI CLI - Offline Privacy-First Commit Generator")
+    print(Fore.CYAN + "                     Developed by PyBloSoft © 2026 - Ver. 1.3.2")
+    print(Fore.CYAN + "═"*86)
     print("Usage:")
     print(f"  gitai         {Style.DIM}Analyze staged changes and generate a conventional commit.{Style.RESET_ALL}")
     print(f"  gitai init    {Style.DIM}Download and configure the local LLM engine weights.{Style.RESET_ALL}")
     print(f"  gitai start   {Style.DIM}Launch the background daemon and prime the prompt cache.{Style.RESET_ALL}")
     print(f"  gitai out     {Style.DIM}Stop the background daemon session and free system RAM.{Style.RESET_ALL}")
     print(f"  gitai help    {Style.DIM}Show this user manual screen.{Style.RESET_ALL}")
-    print(Fore.CYAN + "─"*74)
+    print(Fore.CYAN + "─"*86)
+    print("Privacy & Security Guarantees:")
+    print(f"  • {Fore.GREEN}100% Local Inference:{Style.RESET_ALL} Runs entirely on your CPU/RAM using local GGUF weights.")
+    print(f"  • {Fore.GREEN}Zero Telemetry:{Style.RESET_ALL} No internet connection required, no tracking, no data collection.")
+    print(f"  • {Fore.GREEN}Air-Gapped Ready:{Style.RESET_ALL} Your source code and diffs never leave this machine.")
+    print(Fore.CYAN + "─"*86)
     print("Links & Support:")
     print(f"  • Documentation:  {Fore.BLUE}https://github.com/PabloSalinasDev/GitAI-CLI#readme{Style.RESET_ALL}")
     print(f"  • Report an Issue: {Fore.BLUE}https://github.com/PabloSalinasDev/GitAI-CLI/issues{Style.RESET_ALL}")
-    print(Fore.CYAN + "═"*74 + "\n")
+    print(Fore.CYAN + "═"*86 + "\n")
 
 def main():
     args = sys.argv[1:]
@@ -109,6 +127,10 @@ def main():
 
     # 2. START LOGICAL DAY: gitai start (Check local language and raise the model)
     if args and args[0] == "start":
+        if not is_git_repo():
+            print(Fore.RED + " Error: this directory is not a git repository.")
+            print(Fore.CYAN + " Run 'git init' first.")
+            return
         if not model_exists():
             print(Fore.CYAN + " Model file not found. Run 'gitai init' first to download it.")
             return

{gitai_local-1.3.0 → gitai_local-1.3.2}/gitai_local.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: gitai-local
-Version: 1.3.0
+Version: 1.3.2
 Summary: AI-powered conventional commit message generator running 100% locally
 License-Expression: MIT
 Project-URL: Homepage, https://github.com/PabloSalinasDev/GitAI-CLI
@@ -64,7 +64,7 @@ The benefit of implementing this CLI is twofold:
 3. After staging changes, run `gitai`. It feeds the diff (or initial commit context) into the model and proposes a commit message.
 4. At the end of your session, run `gitai out` to stop the daemon and free RAM.
 
-### 📖 Interactive Help & Documentation Access
+### Interactive Help & Documentation Access
 
 GitAI CLI features a native, robust command-line validation and help module. If you ever need a quick reminder of the available architecture commands, explicit routing links, or want to report a technical issue, you can invoke the user manual directly from your shell.
 
@@ -206,9 +206,9 @@ Based on extensive stress-testing on standard consumer hardware (e.g., Intel i7
 
 | Hardware Profile | Context Size | Estimated Time * | Commit Accuracy |
 |------------------|--------------|------------------|-----------------|
-| **Modern Desktop / Laptop**<br>*(Ryzen 5/7, Core i5/i7 11th+ Gen, DDR5)* | Small to Large Diffs | **~8 – 20 seconds** | ~98% |
-| **Older / Standard Hardware**<br>*(Legacy Intel i7, DDR4 @ 2400MHz)* | Small to Medium Diffs | **~20 – 40 seconds** | ~98% |
-| **Stress Test / Massive Changes**<br>*(Legacy Hardware + Dense Diffs)* | Large Multi-file Diffs | **~40 – 70 seconds** | ~98% |
+| **Modern Desktop / Laptop**<br>*(Ryzen 5/7, Core i5/i7 11th+ Gen, DDR5)* | Small to Large Diffs | **~8 – 20 seconds** | ~90% |
+| **Older / Standard Hardware**<br>*(Legacy Intel i7, DDR4 @ 2400MHz)* | Small to Medium Diffs | **~20 – 40 seconds** | ~90% |
+| **Stress Test / Massive Changes**<br>*(Legacy Hardware + Dense Diffs)* | Large Multi-file Diffs | **~40 – 70 seconds** | ~85% |
 
 > **Note on Performance:** Times are heavily bound to RAM clock speed and single-core efficiency. A dense, multi-file diff analyzed on legacy hardware might occasionally touch the upper boundary of the stress test zone. However, GitAI's built-in traffic manager ensures payloads remain strictly bounded to keep local execution controlled and predictable.
 
@@ -218,7 +218,7 @@ Based on extensive stress-testing on standard consumer hardware (e.g., Intel i7
 
 Why are these times so consistent? GitAI does not just dump raw data into the LLM. It includes a custom pre-processing pipeline that strips out compiler noise, binary files, white spaces, and structural brackets before sending the payload. 
 
-* **Linear vs. Deductive Processing:** Testing proved that sending a complete, clean code block up to 2500 characters is significantly faster than truncating it too early. By providing the model with full, clean context, the LLM processes the data linearly instead of wasting CPU cycles trying to "guess" missing code structures. This balance cuts down processing overhead by up to 30 seconds on older machines while boosting commit accuracy to a staggering **98%**.
+* **Linear vs. Deductive Processing:** Testing proved that sending a complete, clean code block up to 2500 characters is significantly faster than truncating it too early. By providing the model with full, clean context, the LLM processes the data linearly instead of wasting CPU cycles trying to "guess" missing code structures. This balance cuts down processing overhead by up to 40 seconds on older machines while boosting commit accuracy to a staggering **85%**.
 
 ## Dependencies
 
@@ -231,6 +231,33 @@ Why are these times so consistent? GitAI does not just dump raw data into the LL
 
 ---
 
+## Privacy & Security
+
+GitAI is designed with a strict privacy-first architecture:
+
+* **100% Local Execution:** All prompt processing and commit analysis run entirely on your local CPU/GPU using `llama.cpp` bindings.
+* **Zero Telemetry:** The CLI collects no analytics, logs, or usage metrics.
+* **Air-Gapped Friendly:** Once the GGUF model is downloaded, no internet connection is required to generate commits or run commands.
+* **No Code Exposure:** Your repository diffs and code never leave your machine or hit external APIs.
+
+---
+
+## Uninstallation
+
+To completely remove GitAI and delete the downloaded local model weights (~4.7 GB):
+
+```bash
+# 1. Delete the local model cache
+Remove-Item -Recurse -Force "$env:LOCALAPPDATA\gitai"   # Windows
+
+# 2. Uninstall the package
+pipx uninstall gitai-local    # If installed via pipx (README instructions)
+# or
+pip uninstall gitai-local     # If installed via standard pip (PyPI default)
+```
+
+---
+
 ## License
 
 This project is licensed under the MIT License. See the [LICENSE](LICENSE) file for details.

{gitai_local-1.3.0 → gitai_local-1.3.2}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "gitai-local"
-version = "1.3.0"
+version = "1.3.2"
 description = "AI-powered conventional commit message generator running 100% locally"
 readme = "README.md"
 requires-python = ">=3.10"