alpiecode 10.6.9__tar.gz → 10.7.1__tar.gz

{alpiecode-10.6.9 → alpiecode-10.7.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: alpiecode
-Version: 10.6.9
+Version: 10.7.1
 Summary: AlpieCode
 Requires-Python: >=3.9
 Requires-Dist: huggingface_hub>=0.20.0

{alpiecode-10.6.9 → alpiecode-10.7.1}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "alpiecode"
-version = "10.6.9"
+version = "10.7.1"
 description = "AlpieCode"
 requires-python = ">=3.9"
 dependencies = [

{alpiecode-10.6.9 → alpiecode-10.7.1}/src/alpiecode.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: alpiecode
-Version: 10.6.9
+Version: 10.7.1
 Summary: AlpieCode
 Requires-Python: >=3.9
 Requires-Dist: huggingface_hub>=0.20.0

{alpiecode-10.6.9 → alpiecode-10.7.1}/src/codeagent/__init__.py

@@ -17,7 +17,7 @@ Supports:
 from pathlib import Path
 from typing import Optional, Any, Dict
 
-__version__ = "10.6.9"
+__version__ = "10.7.1"
 
 def run(task: str, workdir: str = ".", reasoning_level: str = "thinking", **kwargs):
     """Run an autonomous coding task programmatically."""

{alpiecode-10.6.9 → alpiecode-10.7.1}/src/codeagent/backends/local_backend.py

@@ -877,7 +877,7 @@ class iOSBackend(BasePlatformBackend):
         if tools:
             payload["tools"] = tools
             payload["tool_choice"] = "auto"
-        resp = httpx.post(f"{bridge_url}/v1/chat/completions", json=payload, timeout=90.0)
+        resp = httpx.post(f"{bridge_url}/v1/chat/completions", json=payload, timeout=300.0)
         if resp.status_code != 200:
             raise RuntimeError(f"iOS bridge HTTP {resp.status_code}: {resp.text}")
         data = json.loads(resp.text)
@@ -956,7 +956,7 @@ class iOSBackend(BasePlatformBackend):
         if tools:
             payload["tools"] = tools
             payload["tool_choice"] = "auto"
-        client = httpx.Client(timeout=90.0)
+        client = httpx.Client(timeout=httpx.Timeout(300.0, connect=10.0, read=300.0, write=30.0))
         req = client.build_request("POST", f"{bridge_url}/v1/chat/completions", json=payload)
         resp = client.send(req, stream=True)
         if resp.status_code != 200:
@@ -1007,27 +1007,63 @@ class AndroidBackend(BasePlatformBackend):
     def _get_bridge_url(self) -> str:
         return os.environ.get("ANDROID_LLAMA_BRIDGE", "http://127.0.0.1:8080").rstrip("/")
 
+    def _find_llama_server_bin(self) -> Optional[Path]:
+        custom = os.environ.get("LLAMA_SERVER_PATH") or os.environ.get("LLAMA_BIN")
+        if custom and Path(custom).is_file():
+            return Path(custom)
+        which_bin = shutil.which("llama-server")
+        if which_bin:
+            return Path(which_bin)
+        candidates = [
+            Path("/data/data/com.termux/files/usr/bin/llama-server"),
+            Path.home() / "llama.cpp" / "build" / "bin" / "llama-server",
+            Path.home() / "llama.cpp" / "build" / "bin" / "Release" / "llama-server",
+            Path.home() / ".alpiecode" / "bin" / "llama-server",
+            Path.cwd() / "llama.cpp" / "build" / "bin" / "llama-server",
+        ]
+        for c in candidates:
+            if c.is_file() and os.access(c, os.X_OK):
+                return c
+        return None
+
+    def _find_gguf_model(self) -> Optional[Path]:
+        from ..local_model import get_model_cache_dir
+        candidate_dirs = [
+            get_model_cache_dir(),
+            Path.home() / ".alpiecode" / "models",
+            Path.home() / "models",
+            Path.cwd() / "models",
+            Path.cwd() / ".alpiecode" / "models",
+        ]
+        for d in candidate_dirs:
+            if d.exists():
+                ggufs = list(d.glob("*.gguf"))
+                mains = [g for g in ggufs if "mmproj" not in g.name.lower()]
+                if mains:
+                    mains.sort(key=lambda p: p.stat().st_size)
+                    return mains[0]
+        return None
+
     def _ensure_local_server(self) -> bool:
         """Auto-detect or auto-start local llama-server in background if needed."""
         if self._is_bridge_online():
             return True
-        server_bin = shutil.which("llama-server")
+        server_bin = self._find_llama_server_bin()
         if not server_bin:
+            self._last_server_error = "llama-server binary was not found in PATH or ~/llama.cpp/build/bin/llama-server."
             return False
-        from ..local_model import get_model_cache_dir
-        cache_dir = get_model_cache_dir()
-        if not cache_dir.exists():
-            return False
-        ggufs = list(cache_dir.glob("*.gguf"))
-        main_models = [g for g in ggufs if "mmproj" not in g.name.lower()]
-        if not main_models:
+        model_path = self._find_gguf_model()
+        if not model_path:
+            self._last_server_error = "No GGUF model found in ~/.alpiecode/models or ~/models."
             return False
-        model_path = main_models[0]
-        mmprojs = [g for g in ggufs if "mmproj" in g.name.lower()]
-        cmd = [
-            server_bin,
+        n_ctx = getattr(self._cfg, "n_ctx", 2048)
+        if n_ctx > 2048:
+            n_ctx = 2048
+        n_ctx_val = str(n_ctx)
+        base_cmd = [
+            str(server_bin),
             "-m", str(model_path),
-            "-c", "4096",
+            "-c", n_ctx_val,
             "-t", "4",
             "-np", "1",
             "-b", "512",
@@ -1036,9 +1072,21 @@ class AndroidBackend(BasePlatformBackend):
             "--cache-type-v", "q8_0",
             "--port", "8080",
         ]
-        if mmprojs:
-            cmd += ["--mmproj", str(mmprojs[0])]
+        # Only load vision projector if specifically needed to keep RAM footprint under 5GB
+        if getattr(self._cfg, "enable_vision", False):
+            mmprojs = list(model_path.parent.glob("*mmproj*.gguf"))
+            if mmprojs:
+                base_cmd += ["--mmproj", str(mmprojs[0])]
+
+        taskset_bin = shutil.which("taskset")
+        if taskset_bin and os.cpu_count() == 8:
+            cmd = [taskset_bin, "-c", "4-7"] + base_cmd
+        else:
+            cmd = base_cmd
+
         try:
+            print(f"⏳ Starting Alpie ({model_path.name}) in background...")
+            print(f"   Context window: {n_ctx_val} tokens | Performance cores: 4 | Lean KV cache active")
             subprocess.Popen(
                 cmd,
                 stdout=subprocess.DEVNULL,
@@ -1046,12 +1094,13 @@ class AndroidBackend(BasePlatformBackend):
                 start_new_session=True,
             )
             import time
-            for _ in range(25):
+            for _ in range(40):
                 if self._is_bridge_online():
                     return True
-                time.sleep(0.4)
-        except Exception:
-            pass
+                time.sleep(0.5)
+        except Exception as e:
+            self._last_server_error = f"Failed to launch llama-server: {e}"
+            return False
         return self._is_bridge_online()
 
     @property
@@ -1060,7 +1109,21 @@ class AndroidBackend(BasePlatformBackend):
             import llama_cpp
             return True
         except ImportError:
-            return self._is_bridge_online() or bool(shutil.which("llama-server") or shutil.which("llama-cli"))
+            return self._is_bridge_online() or bool(self._find_llama_server_bin() or shutil.which("llama-cli"))
+
+    def _raise_missing_engine_error(self) -> None:
+        err_detail = getattr(self, "_last_server_error", None)
+        detail_msg = f"\n│  Reason: {err_detail}" if err_detail else ""
+        raise RuntimeError(
+            "\n╭────────────────────────────────────────────────────────────╮\n"
+            "│  Android / Termux Local Inference:                         │\n"
+            "│  Local inference server (http://127.0.0.1:8080) is offline.│"
+            f"{detail_msg}\n"
+            "│                                                            │\n"
+            "│  To start it manually in Termux:                           │\n"
+            "│    llama-server -m ~/.alpiecode/models/*.gguf -c 2048 -t 4 │\n"
+            "╰────────────────────────────────────────────────────────────╯"
+        )
 
     def chat_completion(
         self,
@@ -1079,15 +1142,7 @@ class AndroidBackend(BasePlatformBackend):
         if self._ensure_local_server():
             return self._bridge_chat_completion(messages, tools, temperature, max_tokens, enable_thinking)
 
-        raise RuntimeError(
-            "\n╭────────────────────────────────────────────────────────────╮\n"
-            "│  Android / Termux Native Inference:                        │\n"
-            "│  No local engine is available.                             │\n"
-            "│  To enable local inference on Termux, run:                 │\n"
-            "│    pkg install llama-cpp                                   │\n"
-            "│    alpiecode init                                          │\n"
-            "╰────────────────────────────────────────────────────────────╯"
-        )
+        self._raise_missing_engine_error()
 
     def chat_completion_stream(
         self,
@@ -1108,15 +1163,7 @@ class AndroidBackend(BasePlatformBackend):
             yield from self._bridge_chat_completion_stream(messages, tools, temperature, max_tokens, enable_thinking)
             return
 
-        raise RuntimeError(
-            "\n╭────────────────────────────────────────────────────────────╮\n"
-            "│  Android / Termux Native Inference:                        │\n"
-            "│  No local engine is available.                             │\n"
-            "│  To enable local inference on Termux, run:                 │\n"
-            "│    pkg install llama-cpp                                   │\n"
-            "│    alpiecode init                                          │\n"
-            "╰────────────────────────────────────────────────────────────╯"
-        )
+        self._raise_missing_engine_error()
 
 
 class MacOSBackend(BasePlatformBackend):

{alpiecode-10.6.9 → alpiecode-10.7.1}/src/codeagent/backends/openai_backend.py

@@ -180,7 +180,7 @@ class _HTTPXClient:
         elif "127.0.0.1" in base_url or "localhost" in base_url:
             self.http_client = httpx.Client(
                 http2=False,
-                timeout=httpx.Timeout(180.0, connect=5.0, read=180.0, write=30.0),
+                timeout=httpx.Timeout(300.0, connect=10.0, read=300.0, write=30.0),
                 limits=httpx.Limits(max_keepalive_connections=5, max_connections=10),
             )
         else:

{alpiecode-10.6.9 → alpiecode-10.7.1}/src/codeagent/config.py

@@ -233,9 +233,13 @@ def load_config() -> Config:
         except ValueError:
             pass
 
-    # For offline mode, set high-performance 16k context (16384 tokens) for big multi-file projects
-    # Fits completely in 6GB VRAM on RTX 3050 with zero paging
-    if data.get("base_url") == "offline" and not any(os.environ.get(v) for v in ("ALPIECODE_N_CTX", "ALPIECODE_CTX_SIZE")):
+    # On mobile (Android / Termux), keep context at 2048 to prevent ZRAM thrashing with 9B models
+    if is_android() and not any(os.environ.get(v) for v in ("ALPIECODE_N_CTX", "ALPIECODE_CTX_SIZE")):
+        if data.get("n_ctx", 32768) > 2048:
+            data["n_ctx"] = 2048
+        if data.get("max_tokens", 8192) > 1024:
+            data["max_tokens"] = 1024
+    elif data.get("base_url") == "offline" and not any(os.environ.get(v) for v in ("ALPIECODE_N_CTX", "ALPIECODE_CTX_SIZE")):
         if data.get("n_ctx") in (32768, 4096, 16384):
             data["n_ctx"] = 8192
         if data.get("max_tokens", 8192) > 4096:

{alpiecode-10.6.9 → alpiecode-10.7.1}/test/test_non_thinking_and_local_stream.py

@@ -134,3 +134,28 @@ def test_prompt_builder_lean_offline_prompt():
     assert "You are running in OFFLINE mode" in prompt
     assert "Execution & Goal Convergence Rules" not in prompt
     assert len(prompt) < 3500  # Lean offline prompt under 3500 chars (~600 tokens vs 4500 cloud tokens)
+
+
+def test_android_backend_skips_mmproj_for_code_tasks(tmp_path):
+    from pathlib import Path
+    from codeagent.backends.local_backend import AndroidBackend
+    cfg = Config(base_url="offline")
+    backend = AndroidBackend(cfg)
+
+    fake_model = tmp_path / "alpie-vl-text-q3_k_m.gguf"
+    fake_model.touch()
+    fake_mmproj = tmp_path / "alpie-vl-mmproj-f16.gguf"
+    fake_mmproj.touch()
+
+    with patch.object(backend, "_find_llama_server_bin", return_value=Path("/bin/llama-server")):
+        with patch.object(backend, "_find_gguf_model", return_value=fake_model):
+            with patch("subprocess.Popen") as mock_popen:
+                with patch.object(backend, "_is_bridge_online", side_effect=[False, True]):
+                    backend._ensure_local_server()
+                    assert mock_popen.called
+                    args = mock_popen.call_args[0][0]
+                    # By default (for code tasks), --mmproj must NOT be loaded
+                    assert "--mmproj" not in args
+                    assert "-c" in args
+                    assert "2048" in args
+