bhtool 0.3.1__tar.gz → 0.4.0__tar.gz

{bhtool-0.3.1 → bhtool-0.4.0}/PKG-INFO

@@ -1,13 +1,13 @@
 Metadata-Version: 2.5
 Name: bhtool
-Version: 0.3.1
+Version: 0.4.0
 Summary: Assorted CLI utilities for Bosco’s workflows (dev, media, macOS, LLM helpers).
 Author-email: Bosco Ho <apposite@gmail.com>
 Requires-Python: >=3.13
 Requires-Dist: cyclopts>=4.10.2
-Requires-Dist: microeval>=0.5.0
 Requires-Dist: numpy>=1.26.4
 Requires-Dist: path>=16.14.0
+Requires-Dist: pydantic-ai>=2.55.0
 Requires-Dist: pydash>=8.0.3
 Requires-Dist: python-dotenv>=1.2.2
 Requires-Dist: rich>=15.0.0
@@ -48,7 +48,7 @@ Assorted CLI utilities that are handy in Bosco’s day-to-day work—version bum
 
 ### Movies + LLM
 
-`b movies --service <name>` selects the backend for one run (`openai`, `groq`, `ollama`, `bedrock` — see `b movies --help`). `LLM_SERVICE` sets the default. Set `LLM_SERVICE` in the process environment or in **`bhtool/.env`** (loaded from the installed package directory). Model defaults come from [microeval](https://pypi.org/project/microeval/)’s config for that service.
+`b movies` asks an LLM for normalized names. `--service <name>` selects a provider for one run (`groq`, `google`, `openai`, `anthropic`, `mistral`, `deepseek`, `xai`, `bedrock`, `ollama` — see `b movies --help`). Without it, the provider comes from `LLM_SERVICE`, else from the first API key in the environment, else `ollama`. Set keys in the process environment or in **`bhtool/.env`** (loaded from the installed package directory). Each provider uses one cheap model from `bhtool/llm_services.py` through [pydantic-ai](https://pypi.org/project/pydantic-ai/).
 
 ---
 

{bhtool-0.3.1 → bhtool-0.4.0}/README.md

@@ -32,7 +32,7 @@ Assorted CLI utilities that are handy in Bosco’s day-to-day work—version bum
 
 ### Movies + LLM
 
-`b movies --service <name>` selects the backend for one run (`openai`, `groq`, `ollama`, `bedrock` — see `b movies --help`). `LLM_SERVICE` sets the default. Set `LLM_SERVICE` in the process environment or in **`bhtool/.env`** (loaded from the installed package directory). Model defaults come from [microeval](https://pypi.org/project/microeval/)’s config for that service.
+`b movies` asks an LLM for normalized names. `--service <name>` selects a provider for one run (`groq`, `google`, `openai`, `anthropic`, `mistral`, `deepseek`, `xai`, `bedrock`, `ollama` — see `b movies --help`). Without it, the provider comes from `LLM_SERVICE`, else from the first API key in the environment, else `ollama`. Set keys in the process environment or in **`bhtool/.env`** (loaded from the installed package directory). Each provider uses one cheap model from `bhtool/llm_services.py` through [pydantic-ai](https://pypi.org/project/pydantic-ai/).
 
 ---
 

{bhtool-0.3.1 → bhtool-0.4.0}/bhtool/cli.py

@@ -116,7 +116,7 @@ def movies_cmd(
 
     :param root_dir: Root directory containing movies (positional); default is current working directory.
     :param execute: If true, perform renames; otherwise dry run (table output).
-    :param service: LLM service for normalization; default from LLM_SERVICE, else openai.
+    :param service: LLM service for normalization; default is detected from the environment.
     """
     import bhtool.list_movies as list_movies_mod
 

{bhtool-0.3.1 → bhtool-0.4.0}/bhtool/list_movies.py

@@ -3,7 +3,6 @@
 import asyncio
 import json
 import os
-import re
 import sys
 import textwrap
 
@@ -16,12 +15,19 @@ from rich.panel import Panel
 from rich.table import Table
 from rich.text import Text
 
-from microeval.llm import get_llm_client
+from pydantic import BaseModel, Field
+from pydantic_ai import Agent
+from pydantic_ai.models import infer_model
+
+from bhtool.llm_services import OLLAMA_BASE_URL, detect_service, model_for
 
 _pkg_root = Path(__file__).parent
 
 load_dotenv(_pkg_root / ".env")
 
+# Suppress the pydantic-ai first-run banner in this CLI.
+os.environ.setdefault("PYDANTIC_AI_NO_BANNER", "1")
+
 _MOVIES_BORDER = "cyan"
 _MOVIES_TITLE = "movies"
 
@@ -90,16 +96,10 @@ NORMALIZE_SYSTEM_PROMPT = textwrap.dedent("""
       extension stays unchanged in the output.
     - Preserve the original spelling of the title; only fix casing and
       punctuation.
-    - Return valid JSON only, no markdown or extra text.
 """).strip()
 
 NORMALIZE_USER_PROMPT_TEMPLATE = textwrap.dedent("""
-    Convert these names to normalized form. Return a single JSON object with
-    two keys:
-    - "directories": list of {{"original": "<dir name>",
-      "normalized": "<normalized dir name>"}}
-    - "files": list of {{"original": "<filename with extension>",
-      "normalized": "<normalized base name without extension>"}}
+    Convert these names to normalized form.
 
     Directory names:
     {dir_list}
@@ -109,81 +109,79 @@ NORMALIZE_USER_PROMPT_TEMPLATE = textwrap.dedent("""
 """).strip()
 
 
-def _parse_llm_json_object(result: dict) -> dict | None:
-    text = (result.get("text") or "").strip()
-    if not text:
-        return None
-    fence = re.search(r"```(?:json)?\s*([\s\S]*?)```", text)
-    if fence:
-        text = fence.group(1).strip()
-    try:
-        data = json.loads(text)
-        return data if isinstance(data, dict) else None
-    except json.JSONDecodeError:
-        pass
-    start, end = text.find("{"), text.rfind("}")
-    if start != -1 and end > start:
-        try:
-            data = json.loads(text[start : end + 1])
-            return data if isinstance(data, dict) else None
-        except json.JSONDecodeError:
-            return None
-    return None
+class NameMapping(BaseModel):
+    """One original name and its normalized form."""
+
+    original: str = Field(description="The original directory or file name.")
+    normalized: str = Field(description="The normalized name.")
+
+
+class NormalizedNameMapping(BaseModel):
+    """Normalized names for the directories and files in one root directory."""
+
+    directories: list[NameMapping] = Field(
+        default_factory=list, description="Normalized directory names."
+    )
+    files: list[NameMapping] = Field(
+        default_factory=list, description="Normalized file names."
+    )
 
 
 async def normalize_names_with_llm(
-    dir_names, file_names, service="openai", *, console: Console | None = None
+    dir_names, file_names, service=None, *, console: Console | None = None
 ):
     if not dir_names and not file_names:
         return {"directories": [], "files": []}
 
     console = console or Console()
+    service = service or detect_service()
     dir_list = "\n".join(f"- {n}" for n in dir_names) or "(none)"
     file_list = "\n".join(f"- {n}" for n in file_names) or "(none)"
     user_content = NORMALIZE_USER_PROMPT_TEMPLATE.format(
         dir_list=dir_list,
         file_list=file_list,
     )
-    messages = [
-        {"role": "system", "content": NORMALIZE_SYSTEM_PROMPT},
-        {"role": "user", "content": user_content},
-    ]
 
     try:
-        async with get_llm_client(service) as client:
-            llm_body = Text.assemble(
-                ("service ", ""),
-                (client.service, "bold"),
-                (" model ", ""),
-                (client.model, "bold"),
-                ("\n", ""),
-                ("Requesting normalized names from the LLM…", "dim"),
-            )
-            console.print(_panel_lines(llm_body, title="LLM", border="blue"))
-            result = await client.get_completion(messages)
+        if service == "ollama":
+            os.environ.setdefault("OLLAMA_BASE_URL", OLLAMA_BASE_URL)
+        model = infer_model(model_for(service))
+        agent = Agent(
+            model,
+            instructions=NORMALIZE_SYSTEM_PROMPT,
+            output_type=NormalizedNameMapping,
+        )
+        llm_body = Text.assemble(
+            ("service ", ""),
+            (service, "bold"),
+            (" model ", ""),
+            (model.model_name, "bold"),
+            ("\n", ""),
+            ("Requesting normalized names from the LLM…", "dim"),
+        )
+        console.print(_panel_lines(llm_body, title="LLM", border="blue"))
+        result = await agent.run(user_content)
+        mapping = result.output
     except Exception as e:
         err = Console(stderr=True)
         err.print(CycloptsPanel(str(e), title="Error", style="red"))
         err.print(
             CycloptsPanel(
-                "Set LLM_SERVICE (openai, groq, ollama, bedrock) and the matching API key or "
-                "AWS/Ollama; use bhtool/.env or your environment — see README.",
+                "Set LLM_SERVICE or --service and the matching API key. Providers: "
+                "groq, google, openai, anthropic, mistral, deepseek, xai, bedrock, "
+                "ollama. Use bhtool/.env or your environment — see README.",
                 title="Hint",
                 style="yellow",
             )
         )
         sys.exit(1)
-    if result.get("text", "").strip().startswith("Error:"):
-        raise RuntimeError(result["text"])
 
-    data = _parse_llm_json_object(result)
-    if not data:
-        raise RuntimeError("LLM response could not be parsed as JSON")
-    return {"directories": data.get("directories", []), "files": data.get("files", [])}
+    return mapping.model_dump()
 
 
-def run_normalize_with_llm(root_dir, service="openai", *, console: Console | None = None):
+def run_normalize_with_llm(root_dir, service=None, *, console: Console | None = None):
     console = console or Console()
+    service = service or detect_service()
     root = Path(root_dir)
     video_suffixes = {".avi", ".mkv", ".mp4", ".m4v", ".mov", ".wmv", ".webm"}
     skip_names = {
@@ -219,7 +217,7 @@ def run_normalize_with_llm(root_dir, service="openai", *, console: Console | Non
             f" {'video file' if len(file_names) == 1 else 'video files'} · ",
             "dim",
         ),
-        ("LLM_SERVICE=", "dim"),
+        ("service ", "dim"),
         (service, "bold"),
     )
     console.print(_panel_lines(scan, title="Scan", border=_MOVIES_BORDER))
@@ -364,7 +362,7 @@ def rename(
     )
     console.print(_panel_lines(intro, title=_MOVIES_TITLE, border=_MOVIES_BORDER))
     root = _root_dir(root_dir)
-    svc = service or os.environ.get("LLM_SERVICE", "openai")
+    svc = service or detect_service()
     mapping = run_normalize_with_llm(root, svc, console=console)
     mapping_path = Path.cwd() / "movie_mapping.json"
     mapping_path.write_text(json.dumps(mapping, indent=2))

bhtool-0.4.0/bhtool/llm_services.py

@@ -0,0 +1,77 @@
+#!/usr/bin/env python3
+
+"""Provider defaults for the movies LLM command.
+
+One cheap model per provider, picked for the short name-normalization prompt.
+Prices change, so check them before you add a provider. This module has no
+heavy imports, so the CLI can show the list without loading pydantic-ai.
+"""
+
+import os
+from typing import Literal, get_args
+
+LLMService = Literal[
+    "groq",
+    "google",
+    "openai",
+    "anthropic",
+    "mistral",
+    "deepseek",
+    "xai",
+    "bedrock",
+    "ollama",
+]
+
+PROVIDERS = get_args(LLMService)
+
+# pydantic-ai model id is "<provider>:<model>".
+# The value is the cheapest model that can follow the normalize instructions.
+PROVIDER_MODELS = {
+    "groq": "openai/gpt-oss-20b",         # $0.075 / $0.30 per 1M
+    "google": "gemini-2.5-flash-lite",    # $0.10 / $0.40
+    "openai": "gpt-6-luna",               # $0.10 / $0.50
+    "anthropic": "claude-haiku-5-5",      # $0.10 / $0.50
+    "mistral": "ministral-3-8b-25-12",    # $0.15 / $0.15
+    "deepseek": "deepseek-flash",         # $0.15 / $0.60 off-peak
+    "xai": "grok-4.3",                    # $1.25 / $2.50
+    "bedrock": "amazon.nova-lite-v1:0",   # region-dependent
+    "ollama": "llama3.2",                 # free, local
+}
+
+# Environment variable that enables a provider. Auto-detect reads these in order.
+PROVIDER_ENV_VARS = {
+    "groq": ("GROQ_API_KEY",),
+    "google": ("GOOGLE_API_KEY", "GEMINI_API_KEY"),
+    "openai": ("OPENAI_API_KEY",),
+    "anthropic": ("ANTHROPIC_API_KEY",),
+    "mistral": ("MISTRAL_API_KEY",),
+    "deepseek": ("DEEPSEEK_API_KEY",),
+    "xai": ("XAI_API_KEY",),
+    "bedrock": ("AWS_ACCESS_KEY_ID", "AWS_PROFILE", "AWS_BEARER_TOKEN_BEDROCK"),
+    "ollama": ("OLLAMA_BASE_URL",),
+}
+
+# pydantic-ai needs this value for the Ollama provider.
+OLLAMA_BASE_URL = "http://localhost:11434"
+
+
+def model_for(service: str) -> str:
+    """Return the pydantic-ai model id for a provider, for example 'openai:gpt-6-luna'."""
+    try:
+        model = PROVIDER_MODELS[service]
+    except KeyError:
+        raise ValueError(
+            f"Unknown service {service!r}; choose one of {', '.join(PROVIDERS)}"
+        ) from None
+    return f"{service}:{model}"
+
+
+def detect_service() -> str:
+    """Pick a provider from LLM_SERVICE, else from the environment, else ollama."""
+    env_service = os.environ.get("LLM_SERVICE")
+    if env_service:
+        return env_service.lower()
+    for provider in PROVIDERS:
+        if any(os.environ.get(var) for var in PROVIDER_ENV_VARS[provider]):
+            return provider
+    return "ollama"

{bhtool-0.3.1 → bhtool-0.4.0}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "bhtool"
-version = "0.3.1"
+version = "0.4.0"
 description = "Assorted CLI utilities for Bosco’s workflows (dev, media, macOS, LLM helpers)."
 readme = "README.md"
 requires-python = ">=3.13"
@@ -15,7 +15,7 @@ dependencies = [
     "cyclopts>=4.10.2",
     "python-dotenv>=1.2.2",
     "rich>=15.0.0",
-    "microeval>=0.5.0",
+    "pydantic-ai>=2.55.0",
 ]
 
 [project.scripts]

bhtool-0.3.1/bhtool/llm_services.py

@@ -1,14 +0,0 @@
-#!/usr/bin/env python3
-
-"""LLM service names shared by the CLI and the movies command.
-
-Keep ``LLMService`` in sync with ``microeval.llm.LLM_CLIENTS``. This module has
-no heavy imports, so the CLI can show the choices without importing microeval
-at startup.
-"""
-
-from typing import Literal, get_args
-
-LLMService = Literal["openai", "groq", "ollama", "bedrock"]
-
-LLM_SERVICES = get_args(LLMService)