simplicio-cli 0.5.16__tar.gz → 0.5.17__tar.gz

{simplicio_cli-0.5.16 → simplicio_cli-0.5.17}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: simplicio-cli
-Version: 0.5.16
+Version: 0.5.17
 Summary: Portable task-to-code pipeline that works with any LLM. Turn a one-line task into a verified code change — diff + test + verify loop. +55 pts on a 156-check benchmark, 21% faster, ~same tokens.
 Author-email: Wesley Simplicio <wesleybob4@gmail.com>
 License: MIT
@@ -655,8 +655,7 @@ user prompt. UserPromptSubmit is the right pre-hook for routing decisions.
 | GLM (z.ai) | `glm-4.6` | `https://api.z.ai/api/paas/v4` |
 | DeepSeek | `deepseek-chat` | `https://api.deepseek.com` |
 | OpenAI | `gpt-4.1` | `https://api.openai.com/v1` |
-| Local (Ollama) | `openbmb/minicpm5:latest` | `http://localhost:11434/v1` |
-| Local (in-process) | `local-llama/default` | *(leave unset)* |
+| Local (llama.cpp) | `local-llama/default` | *(leave unset)* |
 | Anthropic native | `claude-opus-4-7` | *(leave unset)* |
 
 If `SIMPLICIO_BASE_URL` is unset and the key is `ANTHROPIC_API_KEY`, it uses the
@@ -667,30 +666,28 @@ your `base_url` — so **any** OpenAI-like provider works without code changes.
 simplicio smoke      # prints provider config + one test call
 ```
 
-#### Path 4 — local Ollama primary with GGUF fallback
+#### Path 4 — local llama.cpp GGUF default
 
 When **no provider is configured** (`SIMPLICIO_MODEL` and
-`SIMPLICIO_BASE_URL` both unset), simplicio uses local Ollama with
-`openbmb/minicpm5:latest`. If that call fails, it falls back to the in-process
+`SIMPLICIO_BASE_URL` both unset), simplicio runs the in-process
 [`llama-cpp-python`](https://github.com/abetlen/llama-cpp-python) backend with
-`Qwen_Qwen3.5-2B-Q6_K.gguf`.
+`local-llama/default`, currently
+`bartowski/Qwen_Qwen3.5-2B-GGUF::Qwen_Qwen3.5-2B-Q6_K.gguf`.
 
 ```bash
 pip install 'simplicio-cli[local]'          # pulls llama-cpp-python + huggingface-hub
+simplicio doctor --install                  # downloads/validates the default GGUF
 
 simplicio task "add input validation to createUser" \
-  --target src/users.ts --local              # forces local Ollama primary
+  --target src/users.ts --local              # forces local llama.cpp
 
-# the fallback GGUF is fetched once from the Hugging Face Hub, then reused
+# the GGUF is fetched once from the Hugging Face Hub, then reused
 ```
 
 Explicit routes (override the default model/weights):
 
 ```bash
-SIMPLICIO_MODEL=openbmb/minicpm5:latest
-SIMPLICIO_BASE_URL=http://localhost:11434/v1
-SIMPLICIO_API_KEY=ollama
-SIMPLICIO_MODEL=local-llama/default                                  # Qwen_Qwen3.5-2B-Q6_K.gguf fallback
+SIMPLICIO_MODEL=local-llama/default                                  # Qwen_Qwen3.5-2B-Q6_K.gguf default
 SIMPLICIO_MODEL=local-llama/bartowski/Qwen_Qwen3.5-2B-GGUF::Qwen_Qwen3.5-2B-Q6_K.gguf
 SIMPLICIO_MODEL=local-llama//models/my-model.gguf                    # direct local path
 SIMPLICIO_LOCAL_MODEL_PATH=/models/my-model.gguf                     # always wins

{simplicio_cli-0.5.16 → simplicio_cli-0.5.17}/README.md

@@ -610,8 +610,7 @@ user prompt. UserPromptSubmit is the right pre-hook for routing decisions.
 | GLM (z.ai) | `glm-4.6` | `https://api.z.ai/api/paas/v4` |
 | DeepSeek | `deepseek-chat` | `https://api.deepseek.com` |
 | OpenAI | `gpt-4.1` | `https://api.openai.com/v1` |
-| Local (Ollama) | `openbmb/minicpm5:latest` | `http://localhost:11434/v1` |
-| Local (in-process) | `local-llama/default` | *(leave unset)* |
+| Local (llama.cpp) | `local-llama/default` | *(leave unset)* |
 | Anthropic native | `claude-opus-4-7` | *(leave unset)* |
 
 If `SIMPLICIO_BASE_URL` is unset and the key is `ANTHROPIC_API_KEY`, it uses the
@@ -622,30 +621,28 @@ your `base_url` — so **any** OpenAI-like provider works without code changes.
 simplicio smoke      # prints provider config + one test call
 ```
 
-#### Path 4 — local Ollama primary with GGUF fallback
+#### Path 4 — local llama.cpp GGUF default
 
 When **no provider is configured** (`SIMPLICIO_MODEL` and
-`SIMPLICIO_BASE_URL` both unset), simplicio uses local Ollama with
-`openbmb/minicpm5:latest`. If that call fails, it falls back to the in-process
+`SIMPLICIO_BASE_URL` both unset), simplicio runs the in-process
 [`llama-cpp-python`](https://github.com/abetlen/llama-cpp-python) backend with
-`Qwen_Qwen3.5-2B-Q6_K.gguf`.
+`local-llama/default`, currently
+`bartowski/Qwen_Qwen3.5-2B-GGUF::Qwen_Qwen3.5-2B-Q6_K.gguf`.
 
 ```bash
 pip install 'simplicio-cli[local]'          # pulls llama-cpp-python + huggingface-hub
+simplicio doctor --install                  # downloads/validates the default GGUF
 
 simplicio task "add input validation to createUser" \
-  --target src/users.ts --local              # forces local Ollama primary
+  --target src/users.ts --local              # forces local llama.cpp
 
-# the fallback GGUF is fetched once from the Hugging Face Hub, then reused
+# the GGUF is fetched once from the Hugging Face Hub, then reused
 ```
 
 Explicit routes (override the default model/weights):
 
 ```bash
-SIMPLICIO_MODEL=openbmb/minicpm5:latest
-SIMPLICIO_BASE_URL=http://localhost:11434/v1
-SIMPLICIO_API_KEY=ollama
-SIMPLICIO_MODEL=local-llama/default                                  # Qwen_Qwen3.5-2B-Q6_K.gguf fallback
+SIMPLICIO_MODEL=local-llama/default                                  # Qwen_Qwen3.5-2B-Q6_K.gguf default
 SIMPLICIO_MODEL=local-llama/bartowski/Qwen_Qwen3.5-2B-GGUF::Qwen_Qwen3.5-2B-Q6_K.gguf
 SIMPLICIO_MODEL=local-llama//models/my-model.gguf                    # direct local path
 SIMPLICIO_LOCAL_MODEL_PATH=/models/my-model.gguf                     # always wins

{simplicio_cli-0.5.16 → simplicio_cli-0.5.17}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "simplicio-cli"
-version = "0.5.16"
+version = "0.5.17"
 description = "Portable task-to-code pipeline that works with any LLM. Turn a one-line task into a verified code change — diff + test + verify loop. +55 pts on a 156-check benchmark, 21% faster, ~same tokens."
 readme = "README.md"
 license = { text = "MIT" }

simplicio_cli-0.5.17/simplicio/__init__.py

@@ -0,0 +1 @@
+__version__ = "0.5.17"

{simplicio_cli-0.5.16 → simplicio_cli-0.5.17}/simplicio/cli.py

@@ -91,8 +91,8 @@ def _add_task_args(p: argparse.ArgumentParser, *, target_required: bool) -> None
     p.add_argument(
         "--local",
         action="store_true",
-        help="force local Ollama (openbmb/minicpm5:latest, no API key) "
-        "with Qwen3.5 GGUF fallback; overrides SIMPLICIO_MODEL/SIMPLICIO_BASE_URL",
+        help="force local llama.cpp with Qwen3.5 GGUF; overrides "
+        "SIMPLICIO_MODEL/SIMPLICIO_BASE_URL",
     )
 
 
@@ -121,11 +121,11 @@ def _add_run_args(p: argparse.ArgumentParser) -> None:
 
 def _force_local_if_requested(a: argparse.Namespace) -> None:
     if getattr(a, "local", False):
-        # Force Path 4: local Ollama primary. The provider layer falls back to
-        # the local Qwen GGUF if the Ollama call fails.
-        os.environ["SIMPLICIO_MODEL"] = "openbmb/minicpm5:latest"
-        os.environ["SIMPLICIO_BASE_URL"] = "http://localhost:11434/v1"
-        os.environ.setdefault("SIMPLICIO_API_KEY", "ollama")
+        # Force Path 4: local in-process llama.cpp. This keeps local execution
+        # independent from Ollama or any HTTP service.
+        os.environ["SIMPLICIO_MODEL"] = "local-llama/default"
+        os.environ.pop("SIMPLICIO_BASE_URL", None)
+        os.environ.pop("SIMPLICIO_API_KEY", None)
 
 
 def _run_task_command(a: argparse.Namespace) -> int:
@@ -609,6 +609,11 @@ def main(argv=None):
     p_status.add_argument("--root", default=".")
     p_status.add_argument("--json", action="store_true")
 
+    p_doctor = sub.add_parser("doctor", help="check local llama.cpp readiness")
+    p_doctor.add_argument("--install", action="store_true")
+    p_doctor.add_argument("--json", action="store_true")
+    p_doctor.add_argument("--list-tiers", action="store_true")
+
     p_env_export = sub.add_parser(
         "env-export",
         help="print shell-safe exports from a dotenv file without sourcing it",
@@ -678,6 +683,17 @@ def main(argv=None):
         return detect_main(detect_argv)
     elif a.cmd == "status":
         return _run_status_command(a)
+    elif a.cmd == "doctor":
+        from .doctor import main as doctor_main
+
+        doctor_argv = []
+        if a.install:
+            doctor_argv.append("--install")
+        if a.json:
+            doctor_argv.append("--json")
+        if a.list_tiers:
+            doctor_argv.append("--list-tiers")
+        return doctor_main(doctor_argv)
     elif a.cmd == "env-export":
         from .runtime_env import parse_env_file, shell_export_lines
 

{simplicio_cli-0.5.16 → simplicio_cli-0.5.17}/simplicio/doctor.py

@@ -1,7 +1,7 @@
 """doctor.py — `simplicio doctor` subcommand.
 
 Prints detected hardware tier + recommended local model + install status.
-With --install, opt-in to pulling the recommended model via Ollama. Without
+With --install, opt-in to downloading the recommended GGUF. Without
 the flag, never touches the disk. With --json, machine-readable output.
 """
 from __future__ import annotations
@@ -14,9 +14,7 @@ from .hardware import detect
 from .local_models import (
     RECOMMENDATIONS,
     ensure_recommended,
-    is_installed,
-    ollama_list_installed,
-    ollama_present,
+    model_file_path,
 )
 
 
@@ -34,46 +32,41 @@ def _render_human(result, profile) -> None:
     print()
     print("recommended doer model:")
     print(f"  label         {result.spec.label}")
-    print(f"  ollama id     {result.spec.ollama_id}")
+    print(f"  model id      {result.spec.model_id}")
+    print(f"  repo          {result.spec.repo_id}")
+    print(f"  file          {result.spec.filename}")
+    print(f"  path          {model_file_path(result.spec)}")
     print(f"  size (Q4)     ~{result.spec.size_gb_q4:.1f} GB")
     print(f"  notes         {result.spec.notes}")
     print()
-    print(f"  ollama        {'installed' if ollama_present() else 'NOT FOUND on PATH'}")
+    print("  runtime       llama.cpp via llama-cpp-python")
     print(f"  can run       {'yes' if result.can_run else 'NO'}")
-    print(f"  can pull      {'yes' if result.can_pull else 'NO'}")
+    print(f"  can download  {'yes' if result.can_download else 'NO'}")
     print(f"  installed     {'YES' if result.installed else 'no'}")
     if result.reason:
         print(f"  status        {result.reason}")
     print()
     if result.installed:
-        print("→ set SIMPLICIO_MODEL to use it:")
-        print(f"  export SIMPLICIO_MODEL={result.spec.ollama_id}")
-        print(f"  export SIMPLICIO_BASE_URL=http://localhost:11434/v1")
-        print(f"  export SIMPLICIO_API_KEY=ollama   # any non-empty string works")
-    elif result.can_pull:
-        print("→ to install:")
+        print("-> set SIMPLICIO_MODEL to use it explicitly:")
+        print(f"  export SIMPLICIO_MODEL={result.spec.model_id}")
+        print("  unset SIMPLICIO_BASE_URL SIMPLICIO_API_KEY")
+    elif result.can_download:
+        print("-> to install:")
         print("  simplicio doctor --install")
-        print(f"  (or manually: ollama pull {result.spec.ollama_id})")
-    elif not ollama_present():
-        print("→ install Ollama first: https://ollama.ai")
+        print(
+            f"  (or manually download {result.spec.repo_id}/{result.spec.filename} "
+            f"to {model_file_path(result.spec)})"
+        )
     else:
-        print("→ hardware is too small for the recommended model; "
+        print("-> hardware is too small for the recommended model; "
               "consider a smaller stack or move to cloud (SIMPLICIO_MODEL = "
               "OpenRouter/HF/etc.)")
 
-    other_installed = [m for m in ollama_list_installed()
-                       if m != result.spec.ollama_id]
-    if other_installed:
-        print()
-        print("other Ollama models you have installed:")
-        for m in other_installed[:10]:
-            print(f"  - {m}")
-
 
 def main(argv: list[str] | None = None) -> int:
     p = argparse.ArgumentParser(prog="simplicio doctor")
     p.add_argument("--install", action="store_true",
-                   help="opt-in: pull the recommended model via Ollama if not present")
+                   help="opt-in: download the recommended GGUF if not present")
     p.add_argument("--json", action="store_true",
                    help="machine-readable output")
     p.add_argument("--list-tiers", action="store_true",
@@ -83,20 +76,26 @@ def main(argv: list[str] | None = None) -> int:
     if args.list_tiers:
         if args.json:
             print(json.dumps({
-                tier: {"ollama_id": s.ollama_id, "label": s.label,
-                       "size_gb_q4": s.size_gb_q4, "notes": s.notes}
+                tier: {
+                    "model_id": s.model_id,
+                    "repo_id": s.repo_id,
+                    "filename": s.filename,
+                    "label": s.label,
+                    "size_gb_q4": s.size_gb_q4,
+                    "notes": s.notes,
+                }
                 for tier, s in RECOMMENDATIONS.items()
             }, indent=2))
         else:
-            print(f"{'tier':14s}  {'size':>7s}  ollama id")
+            print(f"{'tier':14s}  {'size':>7s}  model id")
             print("-" * 80)
             for tier, spec in RECOMMENDATIONS.items():
-                print(f"{tier:14s}  {spec.size_gb_q4:5.1f}GB  {spec.ollama_id}")
-                print(f"  ↳ {spec.label} — {spec.notes}")
+                print(f"{tier:14s}  {spec.size_gb_q4:5.1f}GB  {spec.model_id}")
+                print(f"  -> {spec.label} - {spec.notes}")
         return 0
 
     profile = detect()
-    result = ensure_recommended(profile, auto_pull=args.install)
+    result = ensure_recommended(profile, auto_download=args.install)
 
     if args.json:
         print(json.dumps(result.to_dict(), indent=2))

simplicio_cli-0.5.17/simplicio/local_models.py

@@ -0,0 +1,243 @@
+"""local_models.py - hardware-tier -> llama.cpp GGUF recommendation.
+
+Encodes the local LLM standard:
+
+  all tiers -> local-llama/default
+               bartowski/Qwen_Qwen3.5-2B-GGUF::Qwen_Qwen3.5-2B-Q6_K.gguf
+
+The model runs in-process through llama-cpp-python. No Ollama daemon, pull, or
+HTTP endpoint is required for the default local path.
+
+Hard rule (issue #32 follow-up):
+- NEVER auto-download a model that does not fit the detected tier.
+- Downloads require explicit opt-in (SIMPLICIO_AUTO_DOWNLOAD=1 or
+  `simplicio doctor --install`). We tell the user the command and stop.
+"""
+from __future__ import annotations
+
+import os
+from dataclasses import dataclass
+from pathlib import Path
+
+from .hardware import HardwareProfile
+from .providers import (
+    LOCAL_DEFAULT_FILE as DEFAULT_LOCAL_FILE,
+    LOCAL_DEFAULT_REPO as DEFAULT_LOCAL_REPO,
+    LOCAL_EXECUTOR_DIR,
+    LOCAL_MODEL_PREFIX,
+)
+
+
+DEFAULT_LOCAL_MODEL_ID = f"{LOCAL_MODEL_PREFIX}default"
+DEFAULT_LOCAL_LABEL = "Qwen3.5 2B Q6_K GGUF (llama.cpp)"
+DEFAULT_LOCAL_SIZE_GB = 1.6
+DEFAULT_LOCAL_NOTES = (
+    "canonical local doer; runs in-process with llama-cpp-python, no Ollama service"
+)
+
+
+@dataclass
+class ModelSpec:
+    tier: str
+    model_id: str
+    repo_id: str
+    filename: str
+    size_gb_q4: float
+    label: str
+    notes: str = ""
+
+    @property
+    def ollama_id(self) -> str:
+        """Backward-compatible read alias for older callers."""
+        return self.model_id
+
+
+RECOMMENDATIONS: dict[str, ModelSpec] = {
+    tier: ModelSpec(
+        tier,
+        DEFAULT_LOCAL_MODEL_ID,
+        DEFAULT_LOCAL_REPO,
+        DEFAULT_LOCAL_FILE,
+        DEFAULT_LOCAL_SIZE_GB,
+        DEFAULT_LOCAL_LABEL,
+        DEFAULT_LOCAL_NOTES,
+    )
+    for tier in ("cpu-tiny", "cpu-small", "gpu-mid", "gpu-large", "gpu-xlarge", "unknown")
+}
+
+
+def local_model_dir() -> Path:
+    return Path(os.environ.get("SIMPLICIO_LOCAL_MODEL_DIR", LOCAL_EXECUTOR_DIR)).expanduser()
+
+
+def model_file_path(spec: ModelSpec) -> Path:
+    override = os.environ.get("SIMPLICIO_LOCAL_MODEL_PATH")
+    if override:
+        return Path(override).expanduser()
+    return local_model_dir() / spec.filename
+
+
+def _is_gguf_file(path: Path) -> bool:
+    try:
+        with path.open("rb") as handle:
+            return handle.read(4) == b"GGUF"
+    except OSError:
+        return False
+
+
+def model_file_present(spec: ModelSpec) -> bool:
+    return _is_gguf_file(model_file_path(spec))
+
+
+def is_installed(spec: ModelSpec | str) -> bool:
+    if isinstance(spec, ModelSpec):
+        return model_file_present(spec)
+    if spec.endswith(".gguf"):
+        return _is_gguf_file(local_model_dir() / spec)
+    if spec == DEFAULT_LOCAL_MODEL_ID:
+        return model_file_present(RECOMMENDATIONS["unknown"])
+    return False
+
+
+def download(spec: ModelSpec) -> tuple[bool, str]:
+    """Download the recommended GGUF into the executor model directory."""
+    try:
+        from huggingface_hub import hf_hub_download
+    except ImportError:
+        return (
+            False,
+            "huggingface-hub not installed. Install extras: pip install 'simplicio-cli[local]'",
+        )
+
+    target_dir = local_model_dir()
+    target_dir.mkdir(parents=True, exist_ok=True)
+    try:
+        path = Path(
+            hf_hub_download(
+                repo_id=spec.repo_id,
+                filename=spec.filename,
+                local_dir=str(target_dir),
+            )
+        )
+    except Exception as exc:  # noqa: BLE001 - rendered as CLI status
+        return False, str(exc)
+    if not _is_gguf_file(path):
+        return False, f"{path} is not a valid GGUF file"
+    return True, str(path)
+
+
+@dataclass
+class RecommendationResult:
+    spec: ModelSpec
+    profile: HardwareProfile
+    can_run: bool
+    can_download: bool
+    installed: bool
+    reason: str = ""
+
+    @property
+    def can_pull(self) -> bool:
+        """Backward-compatible read alias for older callers."""
+        return self.can_download
+
+    def to_dict(self) -> dict:
+        return {
+            "tier": self.spec.tier,
+            "model_id": self.spec.model_id,
+            "repo_id": self.spec.repo_id,
+            "filename": self.spec.filename,
+            "model_path": str(model_file_path(self.spec)),
+            "label": self.spec.label,
+            "size_gb_q4": self.spec.size_gb_q4,
+            "notes": self.spec.notes,
+            "ram_gb": round(self.profile.ram_gb, 1),
+            "vram_gb": round(self.profile.vram_gb, 1),
+            "gpu": self.profile.gpu_name,
+            "apple_silicon": self.profile.apple_silicon,
+            "can_run": self.can_run,
+            "can_download": self.can_download,
+            "installed": self.installed,
+            "reason": self.reason,
+        }
+
+
+# Safety margin: refuse to download if the detected resource is not at least
+# (size + margin). We do not want a local automation flow to fill disk/RAM with
+# a model it cannot actually run.
+_SAFETY_MARGIN_GB = 4.0
+
+
+def evaluate(profile: HardwareProfile) -> RecommendationResult:
+    """Pick the local GGUF spec and decide whether the host can run/download it."""
+    spec = RECOMMENDATIONS.get(profile.tier, RECOMMENDATIONS["unknown"])
+
+    if profile.apple_silicon:
+        usable_gb = profile.ram_gb
+    else:
+        usable_gb = max(profile.vram_gb, profile.ram_gb)
+    needed_gb = spec.size_gb_q4 + _SAFETY_MARGIN_GB
+
+    can_run = usable_gb >= needed_gb
+    installed = is_installed(spec)
+    can_download = can_run and not installed
+    reason = ""
+    if not can_run:
+        reason = (
+            f"detected {usable_gb:.1f} GB usable < {needed_gb:.1f} GB "
+            f"required for {spec.label} (size {spec.size_gb_q4:.1f} GB + "
+            f"{_SAFETY_MARGIN_GB:.0f} GB safety margin)"
+        )
+    elif not installed:
+        reason = f"local GGUF not installed at {model_file_path(spec)}"
+
+    return RecommendationResult(
+        spec=spec,
+        profile=profile,
+        can_run=can_run,
+        can_download=can_download,
+        installed=installed,
+        reason=reason,
+    )
+
+
+def ensure_recommended(
+    profile: HardwareProfile,
+    auto_download: bool = False,
+    *,
+    auto_pull: bool | None = None,
+) -> RecommendationResult:
+    """High-level orchestrator for the default local llama.cpp model.
+
+    If the recommended GGUF is missing and auto_download is true, download it.
+    Otherwise return a result the CLI can render so the user knows what to do.
+    `auto_pull` remains as a keyword-only alias for older code paths, but still
+    performs a GGUF download rather than any Ollama action.
+    """
+    if auto_pull is not None:
+        auto_download = auto_download or auto_pull
+
+    result = evaluate(profile)
+    if result.installed:
+        return result
+    if not result.can_download:
+        return result
+
+    do_download = auto_download or os.environ.get(
+        "SIMPLICIO_AUTO_DOWNLOAD", ""
+    ).strip() in ("1", "true", "True", "yes")
+    if not do_download:
+        result.reason = (
+            "model not installed - opt in to download with "
+            "`simplicio doctor --install` or `SIMPLICIO_AUTO_DOWNLOAD=1 ...` "
+            f"(will fetch ~{result.spec.size_gb_q4:.1f} GB)"
+        )
+        return result
+
+    ok, log = download(result.spec)
+    if ok:
+        result.installed = True
+        result.can_download = False
+        result.reason = f"downloaded GGUF to {log}"
+    else:
+        result.reason = f"GGUF download failed: {log[-300:]}"
+    return result