localcode 0.5.27__tar.gz → 0.5.29__tar.gz

{localcode-0.5.27/src/localcode.egg-info → localcode-0.5.29}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: localcode
-Version: 0.5.27
+Version: 0.5.29
 Summary: High-performance AI coding on consumer hardware.
 Author: LocalCode contributors
 License-Expression: Apache-2.0
@@ -65,16 +65,12 @@ Dynamic: license-file
   <img src="https://raw.githubusercontent.com/mjwsolo/localcode/main/docs/assets/demo/first-change.gif" alt="A localcode turn: it reads the stub and test, edits the file, then pytest reports 1 passed" width="900">
 </p>
 
-localcode runs an open-weight model on your Mac and uses it to read, edit and test your code. Model inference runs locally.
-
 ## Install
 
 ```bash
 pip install -U localcode      # or: uv pip install -U localcode
 ```
 
-The inference server ships inside the package. Nothing is compiled or cloned on your machine.
-
 ## Run
 
 ```bash
@@ -82,10 +78,6 @@ cd your-project
 localcode
 ```
 
-```
-> Implement the retry decorator in retry.py so every test in test_retry.py passes. Then run: pytest -q
-```
-
 To use the running model from another local app, open `/server` or run `localcode api` for its
 OpenAI compatible endpoint. See the [Local API guide](https://mjwsolo.github.io/localcode/guides/local-api/).
 
@@ -108,20 +100,18 @@ OpenAI compatible endpoint. See the [Local API guide](https://mjwsolo.github.io/
 
 Choose from **10 models and 100+ quant options** for coding, image input and typed decisions. Models download from Hugging Face. Downloads run in parallel and can resume if interrupted. Switch with `/models`.
 
-| Model | Smallest download | Largest download | 16 GB: tokens/s | 16 GB: context | 128 GB: tokens/s | 128 GB: context |
-| --- | ---: | ---: | ---: | ---: | ---: | ---: |
-| [Gemma 4 12B](https://huggingface.co/unsloth/gemma-4-12b-it-GGUF) | 4.2 GB | 23.8 GB | — | — | — | — |
-| [Qwen 3.8 27B](https://huggingface.co/unsloth/Qwen3.8-27B-GGUF) | 6.2 GB | 31.5 GB | — | — | — | — |
-| [Underdog Saluki 27B 1.0](https://huggingface.co/ConwayResearch/Underdog-Saluki-27B-1.0) | 7.9 GB | 7.9 GB | — | — | — | — |
-| [North-Mini-Code 30B-A3B](https://huggingface.co/unsloth/North-Mini-Code-1.0-GGUF) | 9.4 GB | 33.2 GB | — | — | — | — |
-| [Gemma 4 26B-A4B](https://huggingface.co/unsloth/gemma-4-26B-A4B-it-GGUF) | 9.9 GB | 27.6 GB | — | — | — | — |
-| [Qwen 3.6 35B-A3B](https://huggingface.co/unsloth/Qwen3.6-35B-A3B-GGUF) | 10.0 GB | 38.5 GB | — | — | — | — |
-| [Muse Glimmer 30B](https://huggingface.co/unsloth/Muse-Glimmer-30B-GGUF) | 10.7 GB | 32.3 GB | — | — | — | — |
-| [DiffusionGemma 26B-A4B](https://huggingface.co/unsloth/diffusiongemma-26B-A4B-it-GGUF) | 16.8 GB | 50.5 GB | — | — | — | — |
-| [OpenJev 27B (decisions)](https://huggingface.co/ggml-org/OpenJev-GGUF) | 19.4 GB | 53.8 GB | — | — | — | — |
-| [Kolibri 1 78B-A3B (experimental)](https://huggingface.co/Hob-forge/Kolibri-1-GGUF) | 28.6 GB | 47.5 GB | — | — | — | — |
-
-Download sizes checked on Hugging Face on 12 October 2026; optional vision and draft files are extra. **— = not measured** on that RAM tier. Speed and usable context also depend on the Mac's chip and selected quant; RAM alone does not determine them.
+| Model | Smallest download | Largest download |
+| --- | ---: | ---: |
+| [Gemma 4 12B](https://huggingface.co/unsloth/gemma-4-12b-it-GGUF) | 4.2 GB | 23.8 GB |
+| [Qwen 3.8 27B](https://huggingface.co/unsloth/Qwen3.8-27B-GGUF) | 6.2 GB | 31.5 GB |
+| [Underdog Saluki 27B 1.0](https://huggingface.co/ConwayResearch/Underdog-Saluki-27B-1.0) | 7.9 GB | 7.9 GB |
+| [North-Mini-Code 30B-A3B](https://huggingface.co/unsloth/North-Mini-Code-1.0-GGUF) | 9.4 GB | 33.2 GB |
+| [Gemma 4 26B-A4B](https://huggingface.co/unsloth/gemma-4-26B-A4B-it-GGUF) | 9.9 GB | 27.6 GB |
+| [Qwen 3.6 35B-A3B](https://huggingface.co/unsloth/Qwen3.6-35B-A3B-GGUF) | 10.0 GB | 38.5 GB |
+| [Muse Glimmer 30B](https://huggingface.co/unsloth/Muse-Glimmer-30B-GGUF) | 10.7 GB | 32.3 GB |
+| [DiffusionGemma 26B-A4B](https://huggingface.co/unsloth/diffusiongemma-26B-A4B-it-GGUF) | 16.8 GB | 50.5 GB |
+| [OpenJev 27B (decisions)](https://huggingface.co/ggml-org/OpenJev-GGUF) | 19.4 GB | 53.8 GB |
+| [Kolibri 1 78B-A3B (experimental)](https://huggingface.co/Hob-forge/Kolibri-1-GGUF) | 28.6 GB | 47.5 GB |
 
 ## Network
 

{localcode-0.5.27 → localcode-0.5.29}/README.md

@@ -17,16 +17,12 @@
   <img src="https://raw.githubusercontent.com/mjwsolo/localcode/main/docs/assets/demo/first-change.gif" alt="A localcode turn: it reads the stub and test, edits the file, then pytest reports 1 passed" width="900">
 </p>
 
-localcode runs an open-weight model on your Mac and uses it to read, edit and test your code. Model inference runs locally.
-
 ## Install
 
 ```bash
 pip install -U localcode      # or: uv pip install -U localcode
 ```
 
-The inference server ships inside the package. Nothing is compiled or cloned on your machine.
-
 ## Run
 
 ```bash
@@ -34,10 +30,6 @@ cd your-project
 localcode
 ```
 
-```
-> Implement the retry decorator in retry.py so every test in test_retry.py passes. Then run: pytest -q
-```
-
 To use the running model from another local app, open `/server` or run `localcode api` for its
 OpenAI compatible endpoint. See the [Local API guide](https://mjwsolo.github.io/localcode/guides/local-api/).
 
@@ -60,20 +52,18 @@ OpenAI compatible endpoint. See the [Local API guide](https://mjwsolo.github.io/
 
 Choose from **10 models and 100+ quant options** for coding, image input and typed decisions. Models download from Hugging Face. Downloads run in parallel and can resume if interrupted. Switch with `/models`.
 
-| Model | Smallest download | Largest download | 16 GB: tokens/s | 16 GB: context | 128 GB: tokens/s | 128 GB: context |
-| --- | ---: | ---: | ---: | ---: | ---: | ---: |
-| [Gemma 4 12B](https://huggingface.co/unsloth/gemma-4-12b-it-GGUF) | 4.2 GB | 23.8 GB | — | — | — | — |
-| [Qwen 3.8 27B](https://huggingface.co/unsloth/Qwen3.8-27B-GGUF) | 6.2 GB | 31.5 GB | — | — | — | — |
-| [Underdog Saluki 27B 1.0](https://huggingface.co/ConwayResearch/Underdog-Saluki-27B-1.0) | 7.9 GB | 7.9 GB | — | — | — | — |
-| [North-Mini-Code 30B-A3B](https://huggingface.co/unsloth/North-Mini-Code-1.0-GGUF) | 9.4 GB | 33.2 GB | — | — | — | — |
-| [Gemma 4 26B-A4B](https://huggingface.co/unsloth/gemma-4-26B-A4B-it-GGUF) | 9.9 GB | 27.6 GB | — | — | — | — |
-| [Qwen 3.6 35B-A3B](https://huggingface.co/unsloth/Qwen3.6-35B-A3B-GGUF) | 10.0 GB | 38.5 GB | — | — | — | — |
-| [Muse Glimmer 30B](https://huggingface.co/unsloth/Muse-Glimmer-30B-GGUF) | 10.7 GB | 32.3 GB | — | — | — | — |
-| [DiffusionGemma 26B-A4B](https://huggingface.co/unsloth/diffusiongemma-26B-A4B-it-GGUF) | 16.8 GB | 50.5 GB | — | — | — | — |
-| [OpenJev 27B (decisions)](https://huggingface.co/ggml-org/OpenJev-GGUF) | 19.4 GB | 53.8 GB | — | — | — | — |
-| [Kolibri 1 78B-A3B (experimental)](https://huggingface.co/Hob-forge/Kolibri-1-GGUF) | 28.6 GB | 47.5 GB | — | — | — | — |
-
-Download sizes checked on Hugging Face on 12 October 2026; optional vision and draft files are extra. **— = not measured** on that RAM tier. Speed and usable context also depend on the Mac's chip and selected quant; RAM alone does not determine them.
+| Model | Smallest download | Largest download |
+| --- | ---: | ---: |
+| [Gemma 4 12B](https://huggingface.co/unsloth/gemma-4-12b-it-GGUF) | 4.2 GB | 23.8 GB |
+| [Qwen 3.8 27B](https://huggingface.co/unsloth/Qwen3.8-27B-GGUF) | 6.2 GB | 31.5 GB |
+| [Underdog Saluki 27B 1.0](https://huggingface.co/ConwayResearch/Underdog-Saluki-27B-1.0) | 7.9 GB | 7.9 GB |
+| [North-Mini-Code 30B-A3B](https://huggingface.co/unsloth/North-Mini-Code-1.0-GGUF) | 9.4 GB | 33.2 GB |
+| [Gemma 4 26B-A4B](https://huggingface.co/unsloth/gemma-4-26B-A4B-it-GGUF) | 9.9 GB | 27.6 GB |
+| [Qwen 3.6 35B-A3B](https://huggingface.co/unsloth/Qwen3.6-35B-A3B-GGUF) | 10.0 GB | 38.5 GB |
+| [Muse Glimmer 30B](https://huggingface.co/unsloth/Muse-Glimmer-30B-GGUF) | 10.7 GB | 32.3 GB |
+| [DiffusionGemma 26B-A4B](https://huggingface.co/unsloth/diffusiongemma-26B-A4B-it-GGUF) | 16.8 GB | 50.5 GB |
+| [OpenJev 27B (decisions)](https://huggingface.co/ggml-org/OpenJev-GGUF) | 19.4 GB | 53.8 GB |
+| [Kolibri 1 78B-A3B (experimental)](https://huggingface.co/Hob-forge/Kolibri-1-GGUF) | 28.6 GB | 47.5 GB |
 
 ## Network
 

{localcode-0.5.27 → localcode-0.5.29}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "localcode"
-version = "0.5.27"
+version = "0.5.29"
 description = "High-performance AI coding on consumer hardware."
 readme = "README.md"
 requires-python = ">=3.10"

{localcode-0.5.27 → localcode-0.5.29}/src/localcode/__init__.py

@@ -8,4 +8,4 @@ from importlib.metadata import PackageNotFoundError, version as _pkg_version
 try:
     __version__ = _pkg_version("localcode")
 except PackageNotFoundError:  # not installed (e.g. raw source checkout)
-    __version__ = "0.5.27"
+    __version__ = "0.5.29"

localcode-0.5.29/src/localcode/bin/localcode-ui

@@ -0,0 +1,4 @@
+[diffend] Oversized file quarantined before diffing.
+name: localcode-0.5.29/src/localcode/bin/localcode-ui
+size: 102460976 bytes
+sha256: 5664854fdea46abbdd17d8eab52b61bb75f8a49d3d32af2d82bfbe79b2353cbe

{localcode-0.5.27 → localcode-0.5.29}/src/localcode/bootstrap.py

@@ -344,6 +344,11 @@ def resolve_model_arg(candidate: str | None) -> Path | None:
 
     name = Path(candidate).name
     if name.endswith(".gguf"):
+        explicit = Path(candidate).expanduser()
+        if explicit.is_file():
+            from .models_catalog import by_filename
+            if _is_complete_download(explicit, by_filename(name)):
+                return explicit.resolve()
         return get_model_path(name)
     choice = by_key(candidate)
     if choice is None:
@@ -393,8 +398,9 @@ def get_model_path(preferred_filename: str | None = None) -> Path | None:
     # 3. Canonical download directory (same size check — a half-written file
     # in the canonical dir is a partial, not a usable model).
     if preferred_filename is not None:
-        candidate = _model_dir() / preferred_filename
-        if candidate.is_file() and _is_complete_download(candidate, catalog_entry):
+        from .model_files import find_model
+        candidate = find_model(preferred_filename, _model_dir())
+        if candidate is not None and _is_complete_download(candidate, catalog_entry):
             return candidate
     else:
         # No preference — pick any catalog model that's already downloaded,

{localcode-0.5.27 → localcode-0.5.29}/src/localcode/diagnostics.py

@@ -50,9 +50,41 @@ def report() -> dict:
         'network_configuration_present': {name: bool(os.environ.get(name)) for name in (
             'HTTPS_PROXY', 'HTTP_PROXY', 'https_proxy', 'http_proxy', 'SSL_CERT_FILE',
             'REQUESTS_CA_BUNDLE', 'PIP_INDEX_URL', 'UV_INDEX_URL', 'UV_DEFAULT_INDEX')},
+        'local_services': service_report(),
     }
 
 
+def service_report() -> dict:
+    """Read-only local diagnosis: no API keys, prompts or full process arguments."""
+    from .model_files import model_files
+    from .models_catalog import model_dir
+    from .ui.ports import find_running
+    result = {'context': os.environ.get('LOCALCODE_CONTEXT', 'minimal'),
+              'models_dir': str(model_dir()), 'installed_models': len(model_files(model_dir()))}
+    try:
+        processes = subprocess.check_output(['ps', '-axo', 'pid,ppid,rss,comm'], text=True, timeout=3)
+        servers = []
+        for line in processes.splitlines():
+            fields = line.strip().split(None, 3)
+            if len(fields) == 4 and os.path.basename(fields[3]) == 'llama-server':
+                servers.append({'pid': int(fields[0]), 'parent_pid': int(fields[1]), 'rss_kb': int(fields[2])})
+        result['model_processes'] = servers
+    except (OSError, ValueError, subprocess.SubprocessError):
+        result['model_processes'] = None
+    try:
+        running = find_running()
+        result['supervisor_found'] = running is not None
+        if running:
+            result['control_port'] = running[0]
+            result['server'] = {key: running[1][key] for key in (
+                'version', 'state', 'current', 'port', 'inference_port', 'server_running',
+                'server_ready', 'process_id', 'exit_code', 'ctx', 'warmup',
+            ) if key in running[1]}
+    except (OSError, ValueError):
+        result['supervisor_found'] = False
+    return result
+
+
 def main() -> int:
     print(json.dumps(report(), indent=2))
     return 0

{localcode-0.5.27 → localcode-0.5.29}/src/localcode/entrypoint.py

@@ -16,6 +16,8 @@ def build_parser() -> argparse.ArgumentParser:
     parser.add_argument("--version", action="store_true", help="Print the version and exit")
     parser.add_argument("--mode", choices=["chat", "decisions"], default="chat", help="Start in normal chat or typed decision mode")
     parser.add_argument("--trust-project", action="store_true", help="Allow this project's configuration, plugins and MCP startup commands to load")
+    parser.add_argument("--context", choices=["minimal", "full"], default=os.environ.get("LOCALCODE_CONTEXT", "minimal"),
+                        help="Minimal context by default; full includes project instructions, skills and complete tool history")
     parser.add_argument("--model", help="Model alias to load (otherwise the picker asks)")
     parser.add_argument("--resume", dest="resume", metavar="SESSION_ID",
                         help="Resume a session by ID; uses its saved project directory when no project is given")
@@ -59,6 +61,7 @@ def main(argv: list[str] | None = None) -> None:
 
     args = build_parser().parse_args(argv)
     os.environ["LOCALCODE_MODE"] = args.mode
+    os.environ["LOCALCODE_CONTEXT"] = args.context
 
     if args.version:
         from . import __version__

localcode-0.5.29/src/localcode/model_files.py

@@ -0,0 +1,59 @@
+"""Locate existing GGUF weights without copying or downloading them."""
+from __future__ import annotations
+
+import os
+from pathlib import Path
+
+_SKIP = {'.git', 'node_modules', '.venv', 'venv', '__pycache__', 'dist', 'build', 'voice'}
+
+
+def search_roots(models_dir: Path) -> list[Path]:
+    home = Path.home()
+    roots = [models_dir]
+    try:
+        from .config import load_config
+        configured = (load_config().runtime.model or '').strip()
+        if configured and Path(configured).expanduser().is_file():
+            roots.append(Path(configured).expanduser().parent)
+    except (OSError, ValueError):
+        pass
+    roots.extend([
+        home / '.local/share/localcode/models', home / 'models',
+        Path(os.environ.get('HF_HUB_CACHE') or os.environ.get('HUGGINGFACE_HUB_CACHE') or
+             str(Path(os.environ.get('HF_HOME', home / '.cache/huggingface')) / 'hub')),
+        home / '.lmstudio/models', home / '.cache/lm-studio/models',
+    ])
+    return list(dict.fromkeys(path.expanduser().absolute() for path in roots))
+
+
+def model_files(models_dir: Path) -> dict[str, Path]:
+    """Configured directory wins duplicates; nested snapshots and symlinks work.
+
+    Search only model directories, never the whole user home. Bound traversal
+    and do not follow directory symlinks; file symlinks are ordinary weights.
+    """
+    found: dict[str, Path] = {}
+    directories = 0
+    for root in search_roots(models_dir):
+        if not root.is_dir():
+            continue
+        for directory, children, files in os.walk(root, followlinks=False):
+            directories += 1
+            if directories > 5000:
+                return found
+            depth = len(Path(directory).relative_to(root).parts)
+            children[:] = sorted(name for name in children if name not in _SKIP) if depth < 6 else []
+            for name in sorted(files):
+                if not name.lower().endswith('.gguf') or name.lower().startswith(('mmproj', 'mtp-', 'mtp_', 'dflash')):
+                    continue
+                path = Path(directory) / name
+                if path.is_file():
+                    found.setdefault(name, path)
+    return found
+
+
+def find_model(filename: str, models_dir: Path) -> Path | None:
+    direct = models_dir / filename
+    if direct.is_file():
+        return direct
+    return model_files(models_dir).get(filename)

{localcode-0.5.27 → localcode-0.5.29}/src/localcode/ui/api.py

@@ -5,6 +5,7 @@ import argparse
 import json
 import sys
 import urllib.request
+from localcode.ui.http import local_urlopen
 import urllib.error
 
 from . import run_dir
@@ -39,7 +40,7 @@ def main(argv: list[str] | None = None) -> int:
                 data=json.dumps({"port": port}).encode(),
                 headers={**control_headers(auth[0]), "Content-Type": "application/json"},
             )
-            with urllib.request.urlopen(request, timeout=5) as response:
+            with local_urlopen(request, timeout=5) as response:
                 result = json.load(response)
             print(f"API moved to {result['base_url']}. Update external clients to this address; LocalCode stays connected.")
             return 0

localcode-0.5.29/src/localcode/ui/http.py

@@ -0,0 +1,20 @@
+"""Local service requests never go through an inherited network proxy."""
+from __future__ import annotations
+
+import urllib.request
+from urllib.parse import urlsplit
+
+
+def local_urlopen(request, *, timeout=1):
+    url = request.full_url if isinstance(request, urllib.request.Request) else request
+    if urlsplit(url).hostname not in {"127.0.0.1", "localhost", "::1"}:
+        raise ValueError("Local service requests require a loopback address")
+    return urllib.request.build_opener(urllib.request.ProxyHandler({})).open(request, timeout=timeout)
+
+
+def bypass_local_proxies(env: dict[str, str]) -> None:
+    entries = []
+    for name in ("NO_PROXY", "no_proxy"):
+        entries.extend(part.strip() for part in env.get(name, "").split(",") if part.strip())
+    value = ",".join(dict.fromkeys([*entries, "127.0.0.1", "localhost", "::1"]))
+    env["NO_PROXY"] = env["no_proxy"] = value

{localcode-0.5.27 → localcode-0.5.29}/src/localcode/ui/launch.py

@@ -17,6 +17,7 @@ import subprocess
 import sys
 import time
 import urllib.request
+from localcode.ui.http import bypass_local_proxies, local_urlopen
 from pathlib import Path
 from urllib.parse import quote
 
@@ -27,6 +28,39 @@ from localcode.ui.ports import choose_ports, find_running
 WAIT_S = 240
 
 
+def _shutdown_supervisor(process) -> None:
+    """Allow the guarded model shutdown and kill/reap before forcing exit."""
+    if process.poll() is not None:
+        return
+    process.terminate()
+    try:
+        process.wait(timeout=35)
+    except subprocess.TimeoutExpired:
+        process.kill()
+        process.wait(timeout=5)
+
+
+def _model_alias(model: str | None, directory: Path) -> str | None:
+    if not model:
+        return None
+    from localcode.models_catalog import by_key
+    from localcode.model_files import find_model
+    raw = model.strip()
+    choice = by_key(raw)
+    filename = choice.filename if choice else Path(raw).name
+    if not filename.endswith('.gguf'):
+        filename += '.gguf'
+    explicit = Path(raw).expanduser()
+    existing = explicit if explicit.is_file() else find_model(filename, directory)
+    if existing is None:
+        return None
+    target = directory / existing.name
+    directory.mkdir(parents=True, exist_ok=True)
+    if not target.is_file():
+        target.symlink_to(existing.resolve())
+    return existing.stem
+
+
 def _models_dir() -> Path:
     """Where GGUFs live: LOCALCODE_MODEL_DIR (the documented override, also
     honoured by config), LOCALCODE_MODELS_DIR (the older launcher name), else
@@ -49,7 +83,7 @@ def _get_json(url: str, timeout: float = 1.0) -> dict | None:
     """GET a control-API route with the session token."""
     try:
         req = urllib.request.Request(url, headers=control_headers())
-        with urllib.request.urlopen(req, timeout=timeout) as r:
+        with local_urlopen(req, timeout=timeout) as r:
             return json.load(r)
     except Exception:  # noqa: BLE001
         return None
@@ -58,7 +92,7 @@ def _get_json(url: str, timeout: float = 1.0) -> dict | None:
 def _ok(url: str) -> bool:
     try:
         req = urllib.request.Request(url, headers=control_headers())
-        with urllib.request.urlopen(req, timeout=1.0):
+        with local_urlopen(req, timeout=1.0):
             return True
     except Exception:  # noqa: BLE001
         return False
@@ -167,6 +201,8 @@ def _session_directory(session_id: str) -> Path | None:
 def runtime_env(config_path: Path, ctrl: int, *, trust_project: bool = False) -> dict[str, str]:
     """Project configuration can execute code before tool permissions apply."""
     env = dict(os.environ)
+    env.setdefault("LOCALCODE_CONTEXT", "minimal")
+    bypass_local_proxies(env)
     env["LOCALCODE_CONTROL_URL"] = f"http://127.0.0.1:{ctrl}"
     env["LOCALCODE_CONFIG"] = str(config_path)
     env["OPENCODE_TUI_CONFIG"] = str(config_path.with_suffix(".tui.json"))
@@ -309,11 +345,14 @@ def main(model: str | None = None, project: str | None = None,
               file=sys.stderr)
         return 1
     models_dir = _models_dir()
-    alias = (model or "").strip() or None
-    if alias and not (models_dir / f"{alias}.gguf").is_file():
-        print(f"localcode: no downloaded model named {alias!r} in {models_dir}; opening the picker instead.",
+    try:
+        alias = _model_alias(model, models_dir)
+    except OSError as error:
+        print(f"localcode: could not use the existing model: {error}", file=sys.stderr)
+        return 1
+    if model and alias is None:
+        print(f"localcode: no downloaded model named {model!r} in the model folders; opening the picker instead.",
               file=sys.stderr)
-        alias = None
     project_dir = Path(project).expanduser().resolve() if project else (
         _session_directory(resume) if resume and resume != "last" else None
     ) or Path.cwd()
@@ -322,6 +361,12 @@ def main(model: str | None = None, project: str | None = None,
     # instead of refusing. One llama-server per machine, any number of windows.
     running = find_running()
     if running is not None:
+        from localcode import __version__
+        if running[1].get("version") != __version__:
+            print(f"localcode: a service from an earlier installation is still running. "
+                  f"Exit LocalCode in its open terminals, then launch again to use {__version__}.",
+                  file=sys.stderr)
+            return 1
         return attach(running[0], running[1], ui_bin, project_dir, alias, resume,
                       trust_project=trust_project)
 
@@ -351,17 +396,9 @@ def main(model: str | None = None, project: str | None = None,
     frontend: subprocess.Popen | None = None
 
     def cleanup() -> None:
-        for proc in (frontend, sup):
-            if proc is not None and proc.poll() is None:
-                try:
-                    proc.terminate()
-                except OSError:
-                    pass
-        if sup.poll() is None:
-            try:
-                sup.wait(timeout=8)
-            except subprocess.TimeoutExpired:
-                sup.kill()
+        if frontend is not None and frontend.poll() is None:
+            frontend.terminate()
+        _shutdown_supervisor(sup)
 
     def on_signal(signum, _frame):
         cleanup()

localcode-0.5.29/src/localcode/ui/plugin/context.ts

@@ -0,0 +1,42 @@
+/** Shorten the wire history, preserving the complete saved conversation. */
+const NOTE = "[Earlier tool data abbreviated; reread the relevant file if needed.]";
+
+function brief(value: any, limit: number): any {
+  if (typeof value === "string") {
+    if (value.length <= limit) return value;
+    return value.slice(0, limit - 150) + "\n" + NOTE + "\n" + value.slice(-80);
+  }
+  if (Array.isArray(value)) return value.map((item) => brief(item, limit));
+  if (value && typeof value === "object") {
+    return Object.fromEntries(Object.entries(value).map(([key, item]) => [key,
+      ["filePath", "file_path", "path", "workdir"].includes(key) ? item : brief(item, limit),
+    ]));
+  }
+  return value;
+}
+
+export function leanHistory(messages: any[], workspaceContext = false): any[] {
+  const lastUser = messages.findLastIndex((message) => message.info?.role === "user");
+  return messages.map((message, index) => ({
+    ...message,
+    parts: message.parts.flatMap((part: any) => {
+      if (!workspaceContext && part.synthetic && part.type === "text" &&
+          /^WORKSPACE (SNAPSHOT AT SESSION START|CHANGED SINCE YOUR LAST TURN)/.test(part.text)) return [];
+      // Keep all current-turn data, including tools needed to finish the task.
+      if (index >= lastUser || message.info?.role !== "assistant") return [part];
+      if (part.type === "reasoning") return [];
+      // Explicitly loaded workflows and user decisions remain authoritative.
+      if (part.type !== "tool" || !part.state || ["skill", "question"].includes(part.tool)) return [part];
+      return [{ ...part, state: {
+        ...part.state,
+        input: brief(part.state.input, 600),
+        ...(typeof part.state.output === "string" ? { output: brief(part.state.output, 800) } : {}),
+        ...(typeof part.state.error === "string" ? { error: brief(part.state.error, 400) } : {}),
+        ...(typeof part.state.metadata?.output === "string" ? {
+          metadata: { ...part.state.metadata, output: brief(part.state.metadata.output, 800) },
+        } : {}),
+        ...(part.state.attachments ? { attachments: [] } : {}),
+      } }];
+    }),
+  }));
+}

{localcode-0.5.27 → localcode-0.5.29}/src/localcode/ui/plugin/localcode.ts

@@ -40,6 +40,7 @@ import { appendFileSync, existsSync, mkdirSync, readFileSync, readdirSync, realp
 import { tmpdir } from "node:os";
 import { dirname, isAbsolute, join, relative, resolve, sep } from "node:path";
 import type { Plugin } from "@opencode-ai/plugin";
+import { leanHistory } from "./context";
 
 type Todo = { content: string; status: string };
 
@@ -59,18 +60,13 @@ const controlHeaders = (): Record<string, string> => ({
 const autoCheck = () => process.env.LOCALCODE_AUTO_CHECK === "1";
 
 const PLANNING_RULE = `WORKSPACE TASK COMPLETION:
-- These execution rules apply ONLY when the user asks for work on project files. Answer general questions, advice, greetings, and unrelated web research directly, without todowrite, repository exploration, build checks, or creating files. Do not invent deliverables, compliance requirements, or project work the user did not request.
-- LOCALCODE.md is the only project instruction file and is supplied automatically. Never read, create, or update legacy agent instruction files. Do not seek out configuration or skills for unrelated questions.
-- Do not guess your token count, remaining output budget, or inference speed. Only quote limits or timings supplied by the runtime. For long deliverables, complete the requested work without asking the user to repeat the request.
-- Keep going until the user's request is COMPLETELY done. Do not end your turn while any part of the work remains. A dev server that starts, a scaffold that installs, a single file written — none of these is "done" unless that was the entire request.
-- ORIENT FROM THE SNAPSHOT, NOT FROM TOOL CALLS. The first user turn carries a snapshot of the workspace (top two levels) and later turns report files changed outside this session. Do not list directories or glob just to see what exists; if a file you need is not in the snapshot or a change report, glob for that specific file. Open only the files you will change or must quote. Every tool call is a full model round trip; a typical small change is read -> edit -> check, three calls.
-- PLAN, THEN EXECUTE THE PLAN. For requested workspace changes of THREE or more steps, call todowrite FIRST to lay out every concrete step (one per requirement, and every deliverable the user named — a README is a plan item like any feature, not a closing flourish). Skip the plan for one/two-step tasks. Keep exactly ONE item in_progress. Mark a step completed as soon as its last edit lands (one todowrite call, which may also set the next item in_progress); never call todowrite twice in a row without an edit or check in between, and never before AND after the same edit.
-- ONE EDIT PER FILE PER STEP: when a step changes several places in one file, make them in a single edit or multi-edit call.
-- Validate incrementally: get the minimal scaffold building before implementing the full app. After each coherent feature, run the relevant check and repair failures before adding more features. Do not defer all integration checks until the end.
-- RUN THE CHECK ONCE, AFTER THE EDIT. Do not run tests or builds before you have changed anything unless the user asked for a diagnosis; do not re-run a passing check without a new edit. The last plan item is final verification: the project's own build/typecheck and one smoke check of the main flow. Do not install browsers and test rigs unless requested.
-- Write complete, runnable code — no TODOs, stubs, placeholders, "demo only" or "you could add…". If a piece is too big for one call, split it across calls; never drop it.
-- Only stop for one of two reasons: (a) every todo is completed and verified, or (b) you have ONE specific blocking question you cannot answer yourself. The harness sends you back to the next open item if you stop early.
-- Never run a foreground server (npm run dev, vite, http.server) through bash: start it in the background with nohup ... & and a log file, then curl it.`;
+- Answer greetings, general questions and advice directly. Do not inspect project files, load skills, create todos or run checks unless the user requests project work.
+- For requested project work, use the supplied working directory. Inspect only the relevant files and follow LOCALCODE.md when supplied. Do not seek out legacy agent instructions or unrelated configuration.
+- Complete every requested deliverable and verify it. For three or more steps, make a concise todo list and update it as steps finish. Do not stop at a scaffold or a plan.
+- Match existing project conventions and dependencies. Combine edits to the same file. Run the relevant project check after changes; do not repeat passing checks without new changes. Report failures accurately.
+- Preserve the user's changes. Do not run destructive git commands or commit without permission. Never expose secrets. Background persistent servers and verify their endpoint.
+- Old tool results may be abbreviated in history. Reread a specific file only when needed; never scan the whole project to answer an unrelated question.
+- Use measured limits and timings; do not invent token counts or inference speed. If blocked, state the specific blocker.`;
 
 // A dev server is a PROGRAM being run, so only match in program position (start of
 // command or after ; && || | ( or npx/bunx). Round 5 of the Anki bench died in 2 min
@@ -592,7 +588,9 @@ function workspaceLayout(directory: string): string {
 }
 
 const LocalcodePlugin: Plugin = async ({ client, directory }) => {
-  const layoutBlock = workspaceLayout(directory);
+  // Opt in once at launch; ordinary chat must not scan or attach the project.
+  const workspaceContext = process.env.LOCALCODE_CONTEXT === "full" || process.env.LOCALCODE_WORKSPACE_CONTEXT === "1";
+  const layoutBlock = workspaceContext ? workspaceLayout(directory) : "";
   // Per-session workspace state lives on disk, not in memory: a headless `run`
   // is one process per turn, and a user may close and reopen the UI mid-session.
   // Without this the snapshot was re-sent on every turn and no change was ever reported.
@@ -726,6 +724,12 @@ const LocalcodePlugin: Plugin = async ({ client, directory }) => {
   }
 
   return {
+    "experimental.chat.messages.transform": async (_input, output) => {
+      if (process.env.LOCALCODE_CONTEXT !== "minimal") return;
+      const messages = leanHistory(output.messages, workspaceContext);
+      output.messages.splice(0, output.messages.length, ...messages);
+    },
+
     // Add explicit retention requirements without changing the stable system prefix.
     "experimental.session.compacting": async (_input, output) => {
       if (process.env.LOCALCODE_COMPACTION_CHECKLIST === "0") return;
@@ -784,7 +788,7 @@ const LocalcodePlugin: Plugin = async ({ client, directory }) => {
         turnStartedAt = Date.now() - 1000;
         plateau = new PlateauTracker(directory); plateauStopped = false; stopDenials = 0; stopReason = ""; seenSteps.clear();
       }
-      if (!text.startsWith(NUDGE_PREFIX)) {
+      if (workspaceContext && !text.startsWith(NUDGE_PREFIX)) {
         const st = loadWs(input.sessionID);
         const now = scanTree(directory);
         if (!st || !st.sent) {
@@ -810,7 +814,7 @@ const LocalcodePlugin: Plugin = async ({ client, directory }) => {
       for (const p of editedPaths({ tool: input.tool, args: input.args, output: output?.output, metadata: output?.metadata })) {
         const rel = relative(resolve(directory), resolve(directory, p)).replace(/\\/g, "/");
         agentTouched.add(rel);
-        if (input.sessionID) touchWs(input.sessionID, rel);
+        if (workspaceContext && input.sessionID) touchWs(input.sessionID, rel);
       }
       if (process.env.LOCALCODE_PLUGIN_DEBUG) emit(`[localcode debug] progress: ${plateau.mem.changedPaths.size} delivered files, ${plateau.mem.completedTodos} completed todos`);
     },