devpilot-agentic-cli 1.0.2__py3-none-any.whl → 1.0.4__py3-none-any.whl

agent/cli.py

@@ -12,18 +12,99 @@ import argparse
 import asyncio
 import os
 import sys
+import time
 from datetime import datetime
 from pathlib import Path
 
-from agent.a2a_server import app as a2a_app
+
+def _fix_windows_console() -> None:
+    """
+    Enable ANSI/VT100 support and UTF-8 encoding in Windows terminals.
+
+    Root cause of the blank black CMD screen:
+      - CMD.exe does NOT enable Virtual Terminal Processing by default.
+      - Textual sends ANSI escape sequences to probe terminal capabilities.
+      - Without VT support, CMD either hangs or renders nothing.
+      - PowerShell has VT enabled by default → works immediately.
+
+    Fixes applied:
+      1. Enable ENABLE_VIRTUAL_TERMINAL_PROCESSING via Windows API (ctypes).
+      2. Switch stdout/stderr to UTF-8 so → ✓ etc don't crash in cp1252.
+      3. Set TERM / COLORTERM so Textual skips slow capability negotiation.
+    """
+    if sys.platform != "win32":
+        return
+
+    # ── 1. Enable Virtual Terminal Processing ─────────────────────────────────
+    try:
+        import ctypes
+        import ctypes.wintypes
+        kernel32 = ctypes.windll.kernel32  # type: ignore[attr-defined]
+
+        ENABLE_VIRTUAL_TERMINAL_PROCESSING = 0x0004
+        ENABLE_PROCESSED_OUTPUT            = 0x0001
+
+        for handle_id in (-10, -11):   # STD_INPUT=-10, STD_OUTPUT=-11
+            handle = kernel32.GetStdHandle(handle_id)
+            if handle and handle != -1:
+                mode = ctypes.wintypes.DWORD(0)
+                if kernel32.GetConsoleMode(handle, ctypes.byref(mode)):
+                    new_mode = (mode.value
+                                | ENABLE_VIRTUAL_TERMINAL_PROCESSING
+                                | ENABLE_PROCESSED_OUTPUT)
+                    kernel32.SetConsoleMode(handle, new_mode)
+    except Exception:
+        pass   # non-fatal — worst case CMD is slow, not broken
+
+    # ── 2. Switch stdout/stderr to UTF-8 ──────────────────────────────────────
+    # Prevents UnicodeEncodeError on → ✓ ✗ characters in cp1252 CMD sessions
+    try:
+        if hasattr(sys.stdout, "reconfigure"):
+            sys.stdout.reconfigure(encoding="utf-8", errors="replace")  # type: ignore[union-attr]
+        if hasattr(sys.stderr, "reconfigure"):
+            sys.stderr.reconfigure(encoding="utf-8", errors="replace")  # type: ignore[union-attr]
+        # Also tell Windows console host to use UTF-8 codepage
+        import ctypes
+        ctypes.windll.kernel32.SetConsoleOutputCP(65001)  # type: ignore[attr-defined]
+    except Exception:
+        pass
+
+    # ── 3. Hint terminal type so Textual doesn't hang on capability detection ─
+    # Textual checks these env vars first; without them it sends probe sequences
+    # that CMD ignores, causing a multi-second hang.
+    os.environ.setdefault("TERM", "xterm-256color")
+    os.environ.setdefault("COLORTERM", "truecolor")
+
+
+# Run immediately — must happen before any rich/textual import
+_fix_windows_console()
+
+# Python version guard — must run before any other import
+if sys.version_info < (3, 11):
+    print(
+        f"\n[DevPilot] Python 3.11 or newer is required.\n"
+        f"  You are running Python {sys.version_info.major}.{sys.version_info.minor}.\n"
+        f"  Download the latest Python from: https://python.org/downloads/\n"
+    )
+    sys.exit(1)
+
+# Module-level imports: only lightweight config/ui — everything else is lazy inside main_async()
 from agent.config import Config, ConfigError
-from agent.history import HistoryManager
-from agent.loop import run_agent_loop
-from agent.mcp_client import MCPManager
 from agent.providers.factory import create_provider
-from agent.tools import ToolRegistry
 from agent.ui import UI
 
+_T0 = time.perf_counter()
+
+def _log_startup(msg: str) -> None:
+    """Print a startup progress line. Helps users see DevPilot is loading."""
+    elapsed = time.perf_counter() - _T0
+    # Use plain ASCII fallback so CMD doesn't choke on escape codes if VT failed
+    print(f"\r[{elapsed:.1f}s] {msg}...", end="", flush=True)
+
+def _clear_startup() -> None:
+    """Clear the startup progress line once TUI takes over."""
+    print("\r" + " " * 60 + "\r", end="", flush=True)
+
 
 def parse_args() -> argparse.Namespace:
     parser = argparse.ArgumentParser(
@@ -88,7 +169,8 @@ def _ensure_api_key(config: Config, args: argparse.Namespace) -> Config:
     # Force re-run wizard if --setup flag passed
     if args.setup or not config.active_api_key:
         from agent.setup_wizard import run_setup_wizard
-        success = run_setup_wizard(env_path=Path(".env"))
+        _user_env = Path.home() / ".devpilot" / ".env"
+        success = run_setup_wizard(env_path=_user_env)
         if not success:
             # Wizard skipped (CI/no TTY) or failed — print helpful error
             UI.print_error(
@@ -115,6 +197,7 @@ async def main_async() -> None:
     _apply_cli_overrides(args)
 
     # Load config
+    _log_startup("Loading configuration")
     try:
         config = Config.load()
     except ConfigError as e:
@@ -123,8 +206,11 @@ async def main_async() -> None:
 
     # Handle --setup flag (force wizard even if key exists)
     if args.setup:
+        _clear_startup()
         from agent.setup_wizard import run_setup_wizard
-        run_setup_wizard(env_path=Path(".env"))
+        _user_env = Path.home() / ".devpilot" / ".env"
+        run_setup_wizard(env_path=_user_env)
+
         try:
             config = Config.load()
         except ConfigError as e:
@@ -141,28 +227,36 @@ async def main_async() -> None:
         UI.print_error(str(e))
         sys.exit(1)
 
-    # Wire up provider, context, registry
+    # Wire up provider, context, registry — lazy imports keep startup lean
+    _log_startup("Initializing AI provider")
     provider = create_provider(config)
 
+    _log_startup("Scanning project context")
     from agent.context import RepoContext
+    from agent.history import HistoryManager
+    from agent.tools import ToolRegistry
     repo_context = RepoContext(config.workdir)
     registry = ToolRegistry(config, _context=repo_context)
 
-    # MCP
+    # MCP — lazy import to avoid loading the mcp library at startup
+    _log_startup("Connecting MCP servers")
+    from agent.mcp_client import MCPManager
     mcp_config_path = Path("mcp_servers.json")
     if not mcp_config_path.exists():
         mcp_config_path = Path(__file__).parent.parent / "mcp_servers.json"
     mcp_manager = MCPManager(mcp_config_path)
     await mcp_manager.connect_all(registry)
 
-    # A2A server
-    a2a_app.state.config   = config
-    a2a_app.state.registry = registry
-
+    # A2A server — only import fastapi/uvicorn when A2A is actually enabled
+    _log_startup("Starting A2A server")
     if config.a2a_enabled:
+        from agent.a2a_server import app as a2a_app
         import uvicorn
         import socket
-        
+
+        a2a_app.state.config   = config
+        a2a_app.state.registry = registry
+
         # Check if port is available
         port_in_use = False
         with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
@@ -170,7 +264,7 @@ async def main_async() -> None:
                 s.bind(("0.0.0.0", config.a2a_port))
             except OSError:
                 port_in_use = True
-                
+
         if port_in_use:
             UI.print_error(f"Port {config.a2a_port} is already in use. A2A server disabled. Use --a2a-port <port> to change.")
             a2a_server = None
@@ -207,6 +301,7 @@ async def main_async() -> None:
 
     # ── CI / single-task mode ─────────────────────────────────────────────────
     if args.task:
+        from agent.loop import run_agent_loop
         history.append(provider.make_user_message(args.task))
         await run_agent_loop(
             provider=provider,
@@ -224,7 +319,9 @@ async def main_async() -> None:
         sys.exit(0)
 
     # ── Interactive TUI mode ──────────────────────────────────────────────────
+    _log_startup("Starting DevPilot TUI")
     from agent.tui.app import DevPilotApp
+    _clear_startup()   # erase progress line before TUI takes over the screen
     app = DevPilotApp(
         provider=provider,
         registry=registry,

agent/config.py

@@ -17,8 +17,19 @@ from typing import Literal
 
 from dotenv import load_dotenv
 
-# ── Load .env file if present (safe to call even if file is missing) ─────────
-load_dotenv(dotenv_path=Path(__file__).resolve().parent.parent / ".env", override=False)
+# ── Persistent user config: ~/.devpilot/.env ─────────────────────────────────
+# This is where the setup wizard saves configuration so it persists
+# regardless of which directory `devpilot` is run from.
+_USER_CONFIG_DIR = Path.home() / ".devpilot"
+_USER_ENV_FILE   = _USER_CONFIG_DIR / ".env"
+
+# Load global user config first — this is where the setup wizard saves settings
+load_dotenv(dotenv_path=_USER_ENV_FILE, override=False)
+
+# Load project-local .env second — can ADD vars but won't override global config
+# (override=False means global ~/.devpilot/.env always wins)
+load_dotenv(dotenv_path=Path(".") / ".env", override=False)
+
 
 
 # ── Custom exception ─────────────────────────────────────────────────────────
@@ -32,8 +43,8 @@ class ConfigError(Exception):
 Provider = Literal["anthropic", "openai"]
 
 PROVIDER_DEFAULTS: dict[str, str] = {
-    "anthropic": "claude-opus-4-5",
-    "openai": "gpt-4o",
+    "anthropic": "claude-opus-4-5-20251101",
+    "openai":    "gpt-5.4",
 }
 
 # Keys required per provider (checked lazily so we can build/test without keys)
@@ -43,11 +54,15 @@ REQUIRED_ENV_KEYS: dict[str, str] = {
 }
 
 # Models that support extended thinking (Anthropic only)
+# Verified via Anthropic API — May 2026
 THINKING_CAPABLE_MODELS: set[str] = {
+    "claude-opus-4-7",
+    "claude-opus-4-6",
+    "claude-opus-4-5-20251101",
+    "claude-opus-4-1-20250805",
+    "claude-sonnet-4-5-20250929",
     "claude-3-7-sonnet-20250219",
     "claude-3-5-sonnet-20241022",
-    "claude-opus-4-5",
-    "claude-opus-4-20250514",
 }
 
 

agent/context.py

@@ -145,33 +145,31 @@ class RepoContext:
         return sigs
 
     def _build_project_tree(self, max_entries: int = 200) -> list[str]:
+        """
+        Build a fast file-listing project tree.
+        AST signature extraction is intentionally skipped here to avoid
+        blocking startup — signatures are extracted lazily on file read.
+        """
         entries: list[str] = []
         ignores = {
-            ".git", "node_modules", ".venv", "__pycache__", "dist", 
+            ".git", "node_modules", ".venv", "__pycache__", "dist",
             "build", ".next", ".tox", "coverage_html_report", ".devpilot_sessions"
         }
 
         def _traverse(directory: Path, prefix: str = "") -> None:
             if len(entries) >= max_entries:
                 return
-
             try:
                 for item in sorted(directory.iterdir()):
                     if item.name.startswith(".") and item.name not in (".env", ".gitignore", ".github"):
                         continue
                     if item.name in ignores or item.name.endswith(".egg-info"):
                         continue
-
                     if item.is_dir():
                         entries.append(f"{prefix}📁 {item.name}/")
                         _traverse(item, prefix + "  ")
                     else:
                         entries.append(f"{prefix}📄 {item.name}")
-                        if item.suffix in (".py", ".js", ".ts"):
-                            sigs = self._extract_signatures(item)
-                            for sig in sigs:
-                                entries.append(f"{prefix}{sig}")
-
                     if len(entries) >= max_entries:
                         entries.append("… (truncated to ~200 items for context size)")
                         break

agent/loop.py

@@ -56,8 +56,10 @@ async def run_agent_loop(
         if response.thinking:
             UI.print_thinking_block(response.thinking)
 
-        # UI: print assistant text only when NOT already streamed live.
-        if response.text and not response.streamed_text:
+        # UI: always fire AssistantMessageEvent so the TUI can finalize the response.
+        # For streamed responses the TUI uses the already-buffered text from StreamTokenEvents
+        # and ignores event.text to avoid duplication. For non-streamed it uses event.text directly.
+        if response.text:
             UI.print_assistant_message(response.text)
 
         if not response.has_tool_uses:

agent/mcp_client.py

@@ -8,6 +8,7 @@ and registers them into the ToolRegistry.
 
 import json
 import asyncio
+import shutil
 from pathlib import Path
 
 from mcp.client.session import ClientSession
@@ -18,6 +19,43 @@ from agent.tools import ToolRegistry, ToolResult
 from agent.ui import UI
 
 
+# ── Install hints for common MCP server commands ─────────────────────────────
+
+_INSTALL_HINTS: dict[str, str] = {
+    "npx":    "Node.js is not installed or not on PATH.\n"
+              "  → Install it from https://nodejs.org/ (LTS version recommended)\n"
+              "  → After installing, restart your terminal and try again.",
+    "node":   "Node.js is not installed or not on PATH.\n"
+              "  → Install it from https://nodejs.org/",
+    "uvx":    "'uvx' (uv tool runner) is not installed.\n"
+              "  → Install uv: https://docs.astral.sh/uv/getting-started/installation/",
+    "docker": "Docker is not installed or not running.\n"
+              "  → Install Docker Desktop from https://www.docker.com/products/docker-desktop/",
+    "python": "Python is not on PATH (unexpected — DevPilot itself runs on Python).",
+    "python3":"Python 3 is not on PATH.",
+}
+
+
+def _check_command(command: str, server_name: str) -> bool:
+    """
+    Return True if `command` is available on PATH.
+    If not, print a clear, actionable error and return False.
+    """
+    if shutil.which(command) is not None:
+        return True
+
+    hint = _INSTALL_HINTS.get(command.lower(),
+        f"'{command}' was not found on your PATH.\n"
+        f"  → Make sure it is installed and your terminal can see it."
+    )
+    UI.print_error(
+        f"MCP server '{server_name}' requires '{command}' but it was not found.\n"
+        f"  {hint}\n"
+        f"  This MCP server will be skipped. Disable it in mcp_servers.json to suppress this message."
+    )
+    return False
+
+
 class MCPManager:
     """Manages connections to multiple MCP servers."""
 
@@ -79,7 +117,14 @@ class MCPManager:
             ready_event.set()
 
     async def connect_all(self, registry: ToolRegistry) -> None:
-        """Connect to all servers in mcp_servers.json and register tools."""
+        """Connect to all servers in mcp_servers.json and register tools.
+        
+        Uses a non-blocking approach: MCP servers connect in background tasks.
+        Startup waits at most MCP_CONNECT_TIMEOUT seconds total so a slow
+        npx download never freezes DevPilot at launch.
+        """
+        MCP_CONNECT_TIMEOUT = 2.0  # seconds — enough for cached/local servers; slow npx downloads skip
+
         if not self.config_path.exists():
             return
 
@@ -92,7 +137,7 @@ class MCPManager:
             return
 
         if isinstance(servers, dict):
-            server_items = servers.items()
+            server_items = list(servers.items())
         else:
             server_items = [(s.get("name", f"server_{i}"), s) for i, s in enumerate(servers)]
 
@@ -100,7 +145,7 @@ class MCPManager:
         for name, server_config in server_items:
             if server_config.get("enabled", True) is False:
                 continue
-            
+
             command = server_config.get("command")
             args = server_config.get("args", [])
 
@@ -108,6 +153,10 @@ class MCPManager:
                 UI.print_error(f"MCP server '{name}' missing 'command'. Skipping.")
                 continue
 
+            # Pre-flight: make sure the binary is actually installed
+            if not _check_command(command, name):
+                continue
+
             ready_event = asyncio.Event()
             events.append(ready_event)
             task = asyncio.create_task(
@@ -116,7 +165,21 @@ class MCPManager:
             self._tasks.append(task)
 
         if events:
-            await asyncio.gather(*(e.wait() for e in events))
+            # Wait for all servers to be ready, but cap total wait time.
+            # If any server is slow (e.g. npx downloading a package), DevPilot
+            # still starts promptly and the server connects in the background.
+            try:
+                await asyncio.wait_for(
+                    asyncio.gather(*(e.wait() for e in events)),
+                    timeout=MCP_CONNECT_TIMEOUT,
+                )
+            except asyncio.TimeoutError:
+                pending = sum(1 for e in events if not e.is_set())
+                UI.print_info(
+                    f"{pending} MCP server(s) still connecting in background "
+                    f"(took >{MCP_CONNECT_TIMEOUT:.0f}s). DevPilot is ready now."
+                )
+
 
     async def close(self) -> None:
         """Close all connections."""

agent/setup_wizard.py

@@ -25,24 +25,69 @@ from rich.table import Table
 
 console = Console()
 
-# ── Model lists ───────────────────────────────────────────────────────────────
+
+def _prompt_secret(prompt_text: str) -> str:
+    """
+    Prompt for a secret (API key) showing '*' for each character typed.
+    Works on Windows (CMD / PowerShell) where rich's password=True renders
+    completely blank and provides no visual feedback to the user.
+    Falls back to rich Prompt (password=True) on non-TTY / CI environments.
+    """
+    import sys
+    if not sys.stdin.isatty():
+        return Prompt.ask(prompt_text, password=True).strip()
+
+    import platform
+    if platform.system() == "Windows":
+        import msvcrt
+        console.print(f"{prompt_text}: ", end="")
+        chars: list[str] = []
+        while True:
+            ch = msvcrt.getwch()  # read one char without echo
+            if ch in ("\r", "\n"):   # Enter key
+                console.print()       # move to next line
+                break
+            elif ch == "\x03":        # Ctrl+C
+                raise KeyboardInterrupt
+            elif ch in ("\x08", "\x7f"):  # Backspace
+                if chars:
+                    chars.pop()
+                    # erase last asterisk: backspace + space + backspace
+                    console.print("\b \b", end="")
+            elif ch == "\x00" or ch == "\xe0":  # special key prefix — skip next byte
+                msvcrt.getwch()
+            else:
+                chars.append(ch)
+                console.print("*", end="")
+        return "".join(chars).strip()
+    else:
+        # Unix: getpass already hides input cleanly
+        import getpass
+        return getpass.getpass(f"{prompt_text}: ").strip()
+
+
+# ── Model lists (verified against live APIs) ──────────────────────────────────
+# Last verified: May 2026
 
 _ANTHROPIC_MODELS = [
-    ("claude-opus-4-5-20251101",   "Most capable — best for complex tasks"),
-    ("claude-sonnet-4-5-20251101", "Balanced — fast and capable"),
-    ("claude-haiku-4-5-20251101",  "Fastest — best for simple tasks"),
-    ("claude-3-7-sonnet-20250219", "Extended thinking support"),
+    # Verified via GET https://api.anthropic.com/v1/models
+    ("claude-opus-4-7",            "Newest — most capable Claude (latest)"),
+    ("claude-opus-4-5-20251101",   "Claude Opus 4.5 — powerful & reliable"),
+    ("claude-sonnet-4-5-20250929", "Claude Sonnet 4.5 — balanced speed & quality"),
+    ("claude-haiku-4-5-20251001",  "Claude Haiku 4.5 — fastest, most affordable"),
 ]
 
 _OPENAI_MODELS = [
-    ("gpt-4o",      "Latest GPT-4o — best for coding"),
-    ("gpt-4o-mini", "Fast and cheap — good for simple tasks"),
-    ("o3",          "Most powerful reasoning model"),
-    ("o4-mini",     "Fast reasoning — great for code"),
+    # Verified via platform.openai.com/docs/models (May 2026)
+    # Note: gpt-4o, o3, o4-mini are retired as of early 2026
+    ("gpt-5.5",      "Flagship — best reasoning & coding"),
+    ("gpt-5.4",      "Primary model — coding & professional work"),
+    ("gpt-5.4-mini", "Fast & cost-efficient"),
+    ("gpt-5.4-nano", "Fastest — high-volume low-latency tasks"),
 ]
 
 # OpenAI-compatible third-party providers
-# (base_url, display_name, key_env_var, key_url, models)
+# (display_name, key_url, key_env_var, base_url, models)
 _COMPATIBLE_PROVIDERS = {
     "1": (
         "Groq",
@@ -50,10 +95,12 @@ _COMPATIBLE_PROVIDERS = {
         "GROQ_API_KEY",
         "https://api.groq.com/openai/v1",
         [
-            ("llama-3.3-70b-versatile", "Best Groq model — fast & capable"),
-            ("llama-3.1-8b-instant",    "Ultra-fast, lightweight"),
-            ("mixtral-8x7b-32768",      "Large context window"),
-            ("gemma2-9b-it",            "Google Gemma 2 — fast"),
+            # Verified via GET https://api.groq.com/openai/v1/models
+            ("llama-3.3-70b-versatile",                "Llama 3.3 70B — best quality (131k ctx)"),
+            ("meta-llama/llama-4-scout-17b-16e-instruct","Llama 4 Scout 17B — latest & fast"),
+            ("qwen/qwen3-32b",                         "Qwen3 32B — strong reasoning"),
+            ("llama-3.1-8b-instant",                   "Llama 3.1 8B — ultra-fast"),
+            ("groq/compound",                          "Groq Compound — agentic tasks"),
         ],
     ),
     "2": (
@@ -62,9 +109,12 @@ _COMPATIBLE_PROVIDERS = {
         "TOGETHER_API_KEY",
         "https://api.together.xyz/v1",
         [
-            ("meta-llama/Llama-3-70b-chat-hf", "Llama 3 70B — best quality"),
-            ("meta-llama/Llama-3-8b-chat-hf",  "Llama 3 8B — faster"),
-            ("mistralai/Mixtral-8x7B-v0.1",    "Mixtral — large context"),
+            # Verified via Together AI docs (May 2026)
+            ("meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8", "Llama 4 Maverick — best quality"),
+            ("meta-llama/Llama-4-Scout-17B-16E-Instruct-FP8",     "Llama 4 Scout — fast MoE"),
+            ("meta-llama/Llama-3.3-70B-Instruct-Turbo",           "Llama 3.3 70B Turbo — reliable"),
+            ("meta-llama/Meta-Llama-3.1-405B-Instruct-Turbo",     "Llama 3.1 405B — most capable"),
+            ("meta-llama/Meta-Llama-3.1-8B-Instruct-Turbo",       "Llama 3.1 8B — very fast"),
         ],
     ),
     "3": (
@@ -73,9 +123,10 @@ _COMPATIBLE_PROVIDERS = {
         "MISTRAL_API_KEY",
         "https://api.mistral.ai/v1",
         [
-            ("mistral-large-latest", "Most capable Mistral model"),
-            ("mistral-small-latest", "Fast and affordable"),
-            ("codestral-latest",     "Optimised for code"),
+            # Verified via Mistral docs — latest aliases always point to newest stable
+            ("mistral-large-latest",  "Most capable Mistral model"),
+            ("mistral-small-latest",  "Fast and affordable"),
+            ("codestral-latest",      "Optimised for code generation"),
         ],
     ),
     "4": (
@@ -84,10 +135,11 @@ _COMPATIBLE_PROVIDERS = {
         "OLLAMA_API_KEY",   # Ollama doesn't need a real key
         "http://localhost:11434/v1",
         [
-            ("qwen2.5-coder:7b",  "Best local coding model (recommended)"),
-            ("codellama:13b",     "Meta CodeLlama 13B"),
-            ("deepseek-coder:6b", "DeepSeek Coder 6B"),
-            ("llama3.2:3b",       "Llama 3.2 3B — very fast"),
+            # Run: ollama pull <model> before using
+            ("qwen2.5-coder:7b",    "Qwen2.5-Coder 7B — best local coding (8GB VRAM)"),
+            ("qwen2.5-coder:32b",   "Qwen2.5-Coder 32B — benchmark king (24GB VRAM)"),
+            ("deepseek-coder-v2:16b","DeepSeek-Coder-V2 16B — strong coding (16GB VRAM)"),
+            ("llama3.3:70b",        "Llama 3.3 70B — general purpose"),
         ],
     ),
     "5": (
@@ -100,6 +152,7 @@ _COMPATIBLE_PROVIDERS = {
 }
 
 
+
 # ── Helpers ───────────────────────────────────────────────────────────────────
 
 def _is_interactive() -> bool:
@@ -162,12 +215,17 @@ def _pick_model(models: list[tuple[str, str]], allow_custom: bool = False) -> st
 def run_setup_wizard(env_path: Path | None = None) -> bool:
     """
     Run the interactive first-run setup wizard.
+    Saves configuration to ~/.devpilot/.env so settings persist
+    regardless of which directory `devpilot` is run from.
     Returns True if setup completed, False if skipped/failed.
     """
     if not _is_interactive():
         return False
 
-    env_path = env_path or Path(".env")
+    # Always save to the persistent user config directory
+    user_config_dir = Path.home() / ".devpilot"
+    user_config_dir.mkdir(parents=True, exist_ok=True)
+    env_path = env_path or (user_config_dir / ".env")
 
     console.print(Panel(
         "[bold cyan]Welcome to DevPilot! 🚀[/bold cyan]\n\n"
@@ -179,12 +237,12 @@ def run_setup_wizard(env_path: Path | None = None) -> bool:
 
     # ── Step 1: Choose provider ───────────────────────────────────────────────
     console.print("\n[bold]Step 1 of 3 — Choose your AI provider[/bold]\n")
-    console.print("  [cyan]1[/cyan]  Anthropic      (Claude — recommended)")
-    console.print("  [cyan]2[/cyan]  OpenAI         (GPT-4o, o3, o4-mini)")
-    console.print("  [cyan]3[/cyan]  Groq           (Llama 3, Mixtral — very fast, free tier)")
-    console.print("  [cyan]4[/cyan]  Together AI    (Llama 3, Mixtral)")
-    console.print("  [cyan]5[/cyan]  Mistral AI     (Mistral, Codestral)")
-    console.print("  [cyan]6[/cyan]  Ollama         (local models, no API key needed)")
+    console.print("  [cyan]1[/cyan]  Anthropic      (Claude Opus 4.7, Sonnet 4.5, Haiku 4.5)")
+    console.print("  [cyan]2[/cyan]  OpenAI         (GPT-5.5, GPT-5.4, GPT-5.4-mini)")
+    console.print("  [cyan]3[/cyan]  Groq           (Llama 4 Scout, Llama 3.3 70B — very fast, free tier)")
+    console.print("  [cyan]4[/cyan]  Together AI    (Llama 4 Maverick/Scout, Llama 3.3 70B)")
+    console.print("  [cyan]5[/cyan]  Mistral AI     (Mistral Large, Codestral)")
+    console.print("  [cyan]6[/cyan]  Ollama         (local models — Qwen2.5-Coder, DeepSeek, Llama)")
     console.print("  [cyan]7[/cyan]  Other          (any OpenAI-compatible endpoint)")
 
     choice = Prompt.ask("\nProvider", choices=["1","2","3","4","5","6","7"], default="1")
@@ -195,7 +253,7 @@ def run_setup_wizard(env_path: Path | None = None) -> bool:
     if choice == "1":
         console.print("\n[bold]Step 2 of 3 — Enter your Anthropic API key[/bold]")
         console.print("  Get one at: [link=https://console.anthropic.com/]https://console.anthropic.com/[/link]")
-        api_key = Prompt.ask("\nANTHROPIC_API_KEY", password=True).strip()
+        api_key = _prompt_secret("ANTHROPIC_API_KEY")
         if not api_key:
             console.print("[red]API key cannot be empty.[/red]")
             return False
@@ -215,7 +273,7 @@ def run_setup_wizard(env_path: Path | None = None) -> bool:
     elif choice == "2":
         console.print("\n[bold]Step 2 of 3 — Enter your OpenAI API key[/bold]")
         console.print("  Get one at: [link=https://platform.openai.com/api-keys]https://platform.openai.com/api-keys[/link]")
-        api_key = Prompt.ask("\nOPENAI_API_KEY", password=True).strip()
+        api_key = _prompt_secret("OPENAI_API_KEY")
         if not api_key:
             console.print("[red]API key cannot be empty.[/red]")
             return False
@@ -244,7 +302,7 @@ def run_setup_wizard(env_path: Path | None = None) -> bool:
             api_key = "ollama"
         else:
             console.print(f"  Get one at: [link={key_url}]{key_url}[/link]")
-            api_key = Prompt.ask(f"\n{key_env}", password=True).strip()
+            api_key = _prompt_secret(key_env)
             if not api_key:
                 console.print("[red]API key cannot be empty.[/red]")
                 return False
@@ -264,7 +322,7 @@ def run_setup_wizard(env_path: Path | None = None) -> bool:
     elif choice == "7":
         console.print("\n[bold]Step 2 of 3 — Custom OpenAI-compatible endpoint[/bold]")
         base_url  = Prompt.ask("Base URL (e.g. https://api.example.com/v1)").strip()
-        api_key   = Prompt.ask("API key", password=True).strip()
+        api_key   = _prompt_secret("API key")
         model     = Prompt.ask("Model name (e.g. llama-3-70b)").strip()
 
         if not base_url or not model:

agent/tools/registry.py

@@ -15,15 +15,6 @@ from dataclasses import dataclass, field
 from typing import TYPE_CHECKING, Any, Awaitable, Callable
 
 from agent.tools.base import BaseTool, ToolResult
-from agent.tools.search_code import SearchCodeTool
-
-from agent.tools.shell import RunBashTool
-from agent.tools.a2a import A2ATool
-from agent.tools.web_search import WebSearchTool
-
-from agent.tools.git_ops import GitStatusTool, GitCommitTool
-from agent.tools.doc_gen import DocGenTool
-from agent.tools.diagram import DiagramTool
 from agent.ui import UI
 
 if TYPE_CHECKING:
@@ -108,8 +99,15 @@ class ToolRegistry:
 
     def _register_builtins(self) -> None:
         """Register all default native tools."""
-        # Import here to avoid circular imports at module load time
+        # All heavy tool imports are lazy here — keeps startup fast.
+        # GitPython, tavily, shell, and other deps only load when the
+        # registry is instantiated inside main_async(), not at module import time.
         from agent.tools.fs import ListFilesTool, ReadFileTool, WriteFileTool, EditFileTool
+        from agent.tools.shell import RunBashTool
+        from agent.tools.search_code import SearchCodeTool
+        from agent.tools.git_ops import GitStatusTool, GitCommitTool
+        from agent.tools.doc_gen import DocGenTool
+        from agent.tools.diagram import DiagramTool
 
         tools: list[BaseTool] = [
             ReadFileTool(self._config, self._context),
@@ -118,7 +116,6 @@ class ToolRegistry:
             ListFilesTool(self._config, self._context),
             RunBashTool(self._config),
             SearchCodeTool(self._config),
-
             GitStatusTool(self._config),
             GitCommitTool(self._config),
             DocGenTool(self._config),
@@ -126,10 +123,11 @@ class ToolRegistry:
         ]
 
         if self._config.web_search_enabled:
+            from agent.tools.web_search import WebSearchTool
             tools.append(WebSearchTool(self._config))
 
-
         if self._config.a2a_enabled:
+            from agent.tools.a2a import A2ATool
             tools.append(A2ATool(self._config))
 
         for t in tools:

agent/tui/app.py

@@ -429,22 +429,25 @@ class DevPilotApp(App):
         """Re-enable input when the agent loop finishes."""
         if event.worker.name == "run_agent_task" and event.state.name in ("SUCCESS", "ERROR", "CANCELLED"):
             self.spinner.display = False
-            
+
             try:
                 inp = self.query_one("#chat-input", Input)
                 inp.disabled = False
                 inp.focus()
             except Exception:
                 pass
-            
-            # If we had a buffered stream, write it out
+
+            # Safety fallback: if AssistantMessageEvent never arrived (edge case),
+            # finalize the stream now so the response is never lost.
             if self._active_stream:
                 final_text = self._active_stream._buffer
                 await self._active_stream.remove()
                 self._active_stream = None
-                await self.chat_log.mount(ChatMessage("DevPilot", final_text))
-                self.chat_log.scroll_end(animate=False)
-                self._last_assistant_message = final_text
+                if final_text:
+                    await self.chat_log.mount(ChatMessage("DevPilot", final_text))
+                    self.chat_log.scroll_end(animate=False)
+                    self._last_assistant_message = final_text
+
             # Refresh tree to show newly read/written files
             self._refresh_project_map()
 
@@ -479,14 +482,16 @@ class DevPilotApp(App):
     @on(ThinkingEvent)
     async def handle_ui_events(self, event: UIEvent) -> None:
         if isinstance(event, AssistantMessageEvent):
-            final_text = ""
             if self._active_stream:
+                # Streaming was used — use the already-buffered text.
+                # Ignore event.text to avoid duplication (it's the same content).
                 final_text = self._active_stream._buffer
                 await self._active_stream.remove()
                 self._active_stream = None
-            if event.text.strip():
-                final_text += ("\n" + event.text.strip()) if final_text else event.text.strip()
-                
+            else:
+                # Non-streaming path — event.text is the only source.
+                final_text = event.text.strip()
+
             if final_text:
                 await self.chat_log.mount(ChatMessage("DevPilot", final_text))
                 self.chat_log.scroll_end(animate=False)

{devpilot_agentic_cli-1.0.2.dist-info → devpilot_agentic_cli-1.0.4.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: devpilot-agentic-cli
-Version: 1.0.2
+Version: 1.0.4
 Summary: Autonomous AI coding agent for your terminal — Claude, GPT-4o, Groq, Ollama, and more
 Author: Thijesh Praveen V
 License: MIT
@@ -13,6 +13,8 @@ Classifier: Environment :: Console
 Classifier: Intended Audience :: Developers
 Classifier: Programming Language :: Python :: 3.11
 Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Programming Language :: Python :: 3.14
 Classifier: License :: OSI Approved :: MIT License
 Classifier: Topic :: Software Development
 Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
@@ -46,7 +48,7 @@ Requires-Dist: pytest-asyncio>=0.23.7; extra == "dev"
 [![Python](https://img.shields.io/pypi/pyversions/devpilot-agentic-cli)](https://pypi.org/project/devpilot-agentic-cli/)
 [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
 
-**An autonomous AI coding agent for your terminal.** DevPilot gives Claude, GPT-4o, Groq, Mistral, or any local model a full suite of tools — file read/write, bash execution, code search, git, MCP, and more — orchestrated into a self-healing agentic loop inside a rich Textual TUI.
+**An autonomous AI coding agent for your terminal.** DevPilot gives Claude, GPT, Groq, Mistral, or any local model a full suite of tools — file read/write, bash execution, code search, git, MCP, and more — orchestrated into a self-healing agentic loop inside a rich Textual TUI. Works in PowerShell, CMD, and any modern terminal.
 
 ```
 ┌─ DevPilot ──────────────────────────────────────────────────────┐
@@ -76,7 +78,6 @@ Requires-Dist: pytest-asyncio>=0.23.7; extra == "dev"
 | **Git** | `git_status`, `git_commit` with surgical file staging |
 | **Documentation** | `doc_gen` (markdown), `diagram` (Mermaid) |
 | **Web search** | `web_search` via Tavily (optional) |
-
 | **MCP** | Connect any MCP server via `mcp_servers.json` |
 | **A2A** | Agent-to-agent task delegation over HTTP |
 | **Providers** | Anthropic, OpenAI, Groq, Together AI, Mistral, Ollama, any OpenAI-compatible endpoint |
@@ -92,7 +93,7 @@ pip install devpilot-agentic-cli
 devpilot
 ```
 
-That's it. On first run DevPilot launches a setup wizard — pick your provider, enter your API key, choose a model. It saves everything to a `.env` file and opens the TUI. Every subsequent run starts immediately.
+That's it. On first run DevPilot launches a setup wizard — pick your provider, enter your API key, choose a model. It saves everything to `~/.devpilot/.env` and opens the TUI. Every subsequent run starts immediately.
 
 ### Optional Features
 
@@ -111,19 +112,19 @@ DevPilot works with any of these out of the box — the setup wizard walks you t
 
 | Provider | Models | Get API key |
 |---|---|---|
-| **Anthropic** | claude-opus-4-5, claude-sonnet-4-5, claude-haiku-4-5 | [console.anthropic.com](https://console.anthropic.com/) |
-| **OpenAI** | gpt-4o, gpt-4o-mini, o3, o4-mini | [platform.openai.com](https://platform.openai.com/api-keys) |
-| **Groq** | llama-3.3-70b, mixtral-8x7b, gemma2-9b | [console.groq.com](https://console.groq.com/keys) |
-| **Together AI** | Llama 3, Mixtral, and more | [api.together.xyz](https://api.together.xyz/settings/api-keys) |
-| **Mistral AI** | mistral-large, codestral | [console.mistral.ai](https://console.mistral.ai/api-keys/) |
-| **Ollama** | Any local model — no API key needed | [ollama.com](https://ollama.com/library) |
+| **Anthropic** | claude-opus-4-7, claude-opus-4-5-20251101, claude-sonnet-4-5-20250929, claude-haiku-4-5-20251001 | [console.anthropic.com](https://console.anthropic.com/) |
+| **OpenAI** | gpt-5.5, gpt-5.4, gpt-5.4-mini, gpt-5.4-nano | [platform.openai.com](https://platform.openai.com/api-keys) |
+| **Groq** | llama-3.3-70b-versatile, llama-4-scout-17b, qwen3-32b | [console.groq.com](https://console.groq.com/keys) |
+| **Together AI** | Llama 4 Maverick/Scout, Llama 3.3 70B Turbo | [api.together.xyz](https://api.together.xyz/settings/api-keys) |
+| **Mistral AI** | mistral-large-latest, mistral-small-latest, codestral-latest | [console.mistral.ai](https://console.mistral.ai/api-keys/) |
+| **Ollama** | qwen2.5-coder:7b, deepseek-coder-v2:16b, llama3.3:70b — no API key needed | [ollama.com](https://ollama.com/library) |
 | **Other** | Any OpenAI-compatible endpoint | — |
 
 ---
 
 ## Configuration
 
-All settings live in a `.env` file in your project directory. The setup wizard creates this on first run. You can edit it manually anytime — or re-run the wizard with `devpilot --setup`.
+All settings live in `~/.devpilot/.env` — a persistent file in your home directory that works regardless of which folder you run `devpilot` from. The setup wizard creates this on first run. You can edit it manually anytime — or re-run the wizard with `devpilot --setup`.
 
 ### Full settings reference
 
@@ -131,9 +132,10 @@ All settings live in a `.env` file in your project directory. The setup wizard c
 |---|---|---|
 | `ANTHROPIC_API_KEY` | — | API key for Anthropic provider |
 | `OPENAI_API_KEY` | — | API key for OpenAI / Groq / Together / Mistral / Ollama |
+| `GROQ_API_KEY` | — | API key for Groq (optional — can reuse as OPENAI_API_KEY) |
 | `DEVPILOT_PROVIDER` | `anthropic` | `anthropic` or `openai` |
-| `DEVPILOT_MODEL` | `claude-opus-4-5` | Model name |
-| `DEVPILOT_BASE_URL` | — | Custom endpoint, e.g. `http://localhost:11434/v1` for Ollama |
+| `DEVPILOT_MODEL` | `claude-opus-4-5-20251101` | Model name |
+| `DEVPILOT_BASE_URL` | — | Custom endpoint, e.g. `https://api.groq.com/openai/v1` |
 | `DEVPILOT_NO_CONFIRM` | `false` | Skip confirmation prompts (useful for CI) |
 | `DEVPILOT_MAX_ITERATIONS` | `50` | Max tool-use iterations before loop aborts |
 | `DEVPILOT_WORKDIR` | `cwd` | Root directory for file operations |
@@ -144,12 +146,12 @@ All settings live in a `.env` file in your project directory. The setup wizard c
 ### Override priority
 
 ```
-CLI flags  →  env vars  →  .env file  →  defaults
+CLI flags  →  env vars  →  ~/.devpilot/.env  →  defaults
 ```
 
 Per-run override example:
 ```bash
-DEVPILOT_MODEL=claude-haiku-4-5-20251101 devpilot --task "fix typos"
+DEVPILOT_MODEL=llama-3.3-70b-versatile devpilot --task "fix typos"
 ```
 
 ---
@@ -172,7 +174,6 @@ Options:
   --a2a-port PORT                 A2A server port (default: 8000)
   --no-a2a                        Disable A2A server
   --no-web-search                 Disable Tavily web search
-
   --setup                         Re-run the setup wizard
 ```
 
@@ -232,7 +233,7 @@ agent/
 ├── setup_wizard.py       First-run interactive configuration wizard
 ├── loop.py               Core agentic loop (plan → act → verify → heal)
 ├── config.py             Config dataclass — all settings from env vars
-├── context.py            RepoContext — file awareness, AST map
+├── context.py            RepoContext — file awareness, AST project map
 ├── history.py            Conversation history + smart context pruning
 ├── providers/
 │   ├── anthropic_provider.py
@@ -247,7 +248,6 @@ agent/
 │   ├── doc_gen.py        doc_gen
 │   ├── diagram.py        diagram (Mermaid)
 │   ├── web_search.py     web_search (Tavily)
-
 │   ├── a2a.py            A2A delegation
 │   └── registry.py       ToolRegistry + PermissionGuard
 └── tui/

{devpilot_agentic_cli-1.0.2.dist-info → devpilot_agentic_cli-1.0.4.dist-info}/RECORD

@@ -1,13 +1,13 @@
 agent/__init__.py,sha256=lfJPQT9NbRRWfa9YABD9u8apcLjs3tzDVZvrOb69GBQ,25
 agent/a2a_client.py,sha256=vY4qZahkWP-DVU2smIeTz1xLj61kkDjE9Tmt_-9PKUU,3416
 agent/a2a_server.py,sha256=3rkDy8dm49_5qCa6fKYkLdmHTnHs5ocHBuZ8j32-p48,5230
-agent/cli.py,sha256=YdgLLIk4vgK-gGZ-Ia-kSZ_BbgfB2WE9QX83jiGsv48,8819
-agent/config.py,sha256=GWS719hDNvvzgJAq9MF2FAGEVwM1AKA_B4WSDDzKEAg,10580
-agent/context.py,sha256=a--jOYLX5jM-8RDLV0tAvZHOx9H5M6SNxSJipzEUZe4,7323
+agent/cli.py,sha256=ApAnD7pQUvHPsnHt5g8JoK0CaGAIPS78qiGQOt7B4X4,13305
+agent/config.py,sha256=b53O7kkdAUYTNCm7YLxBYiySj6ZXn2qzoWqgPN1NIUw,11244
+agent/context.py,sha256=atGJgIiMAmYfcNu6k9gwUzsz5cykBAx9A8PCdgK64R0,7297
 agent/history.py,sha256=IMY3IO5TN9Sh7tJPAWE9VZYC0pkAwgPj2a_F1yA4OmE,6730
-agent/loop.py,sha256=8lTvCSAljPdmyyJ1uf0Q9qulUFNE3d4v3ZNzRae4c1w,3514
-agent/mcp_client.py,sha256=lasmMq602_OEirzTNO57dNEF7mWaed_A6NaNTY3voOQ,4978
-agent/setup_wizard.py,sha256=jCXpp1vY9HRqJLO9eC6M9T6JlHXP7LsausmhkciB0d8,12774
+agent/loop.py,sha256=3QxOb2c09ewAcIbAnd3RVCsXzKaGTJFBLZ-5OdhnOi0,3693
+agent/mcp_client.py,sha256=A8WgS2DhfBbpZZGrJYwOZwMAF0gA-YG4gylBgMzkriE,7785
+agent/setup_wizard.py,sha256=KNDxy-0jC6erB9IGpZoPCcrgpzU523D8p1zEx8aMbC0,15655
 agent/ui.py,sha256=LK8L2fTq_JIcYiVVjpVWaoYMFWXwqUcqYL4gspdAqvs,8283
 agent/providers/__init__.py,sha256=6xBg26vCBV2e3hybMT2JCHEZoRNpImkVOK-FytBW4xY,166
 agent/providers/anthropic_provider.py,sha256=i257T8T7fYuUEVvsy-3_eudbjSlxZqbn3zi7DFvukRE,6067
@@ -22,14 +22,14 @@ agent/tools/diagram.py,sha256=FpQRWud6GMi34ysjbEkqnvy5aaHOJwIvpmDELsyIqPE,4884
 agent/tools/doc_gen.py,sha256=TaLw5JJ1hSgfViGyfYRMfUafjRTfxa0JYtZw0ycluR8,6064
 agent/tools/fs.py,sha256=nBrGzq-yIzFcRzhPguBxJ9H23lhsUjGPzQXhwfrcQY0,16121
 agent/tools/git_ops.py,sha256=CbulyzzSuFUQdAqu-6eMkg7CpB45h7nFP_5fxTecHeE,5505
-agent/tools/registry.py,sha256=R6PsHIVqWyJwuFdC2EIm3GLdtTCr4j2S5BGWOTPanNA,8404
+agent/tools/registry.py,sha256=0I8Pgt7obgNaVB8BTeHHRmdouTj3P3Kk7GCCra7D2fg,8620
 agent/tools/search_code.py,sha256=DwxT6D10OpWvIX7KR3-cRW1Z1MkLW2_WcJ47H_jfuHM,4490
 agent/tools/shell.py,sha256=i9nar3gyS28p6wjfAljNREUll-7RiD_LhQQ_CDkz-CM,3923
 agent/tools/web_search.py,sha256=7a533lYcRG0rwYle00xsIncfPRL1zCiS-wsL_qVxlyI,3736
 agent/tui/__init__.py,sha256=FB4u1BY9RSmYVfCM2yMeCyMmzjnb9T0bhmPp9hHH8Ys,37
-agent/tui/app.py,sha256=ZClXMLgXvohtLlYogKMpQISjEiFTHzNP6XPSeM6y4xU,18598
-devpilot_agentic_cli-1.0.2.dist-info/METADATA,sha256=HGoT4GuEaEX-l_-Fp4Cu5sewyRn9X058Bqp7HHUdXlA,11105
-devpilot_agentic_cli-1.0.2.dist-info/WHEEL,sha256=aeYiig01lYGDzBgS8HxWXOg3uV61G9ijOsup-k9o1sk,91
-devpilot_agentic_cli-1.0.2.dist-info/entry_points.txt,sha256=Zi0zH4sqSQptehekvn5CBhU452GQ23Vyja4l8C58IuY,44
-devpilot_agentic_cli-1.0.2.dist-info/top_level.txt,sha256=0gvCG7PHc22NA63j3bTGi2Zc37ym9t8Pf90ZLzf1kGA,6
-devpilot_agentic_cli-1.0.2.dist-info/RECORD,,
+agent/tui/app.py,sha256=dBa3l5J3dkS7j56b-NgTQXR1At8aRXpayORk4CbJGBo,18831
+devpilot_agentic_cli-1.0.4.dist-info/METADATA,sha256=VHAXdhDQOcW-VLetIITT-AZmXmUbsdViyjK7dho1IZE,11608
+devpilot_agentic_cli-1.0.4.dist-info/WHEEL,sha256=aeYiig01lYGDzBgS8HxWXOg3uV61G9ijOsup-k9o1sk,91
+devpilot_agentic_cli-1.0.4.dist-info/entry_points.txt,sha256=Zi0zH4sqSQptehekvn5CBhU452GQ23Vyja4l8C58IuY,44
+devpilot_agentic_cli-1.0.4.dist-info/top_level.txt,sha256=0gvCG7PHc22NA63j3bTGi2Zc37ym9t8Pf90ZLzf1kGA,6
+devpilot_agentic_cli-1.0.4.dist-info/RECORD,,