deepy-cli 0.1.5__tar.gz → 0.1.7__tar.gz

{deepy_cli-0.1.5 → deepy_cli-0.1.7}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: deepy-cli
-Version: 0.1.5
+Version: 0.1.7
 Summary: Deepy - Vibe coding for DeepSeek models in your terminal
 Keywords: deepseek,coding-agent,terminal,cli,agents
 Author: kirineko
@@ -62,7 +62,8 @@ context state visible while the agent works.
 ## Highlights
 
 - DeepSeek-first model setup with `deepseek-v4-pro`, thinking enabled, and
-  `reasoning_effort=max` by default.
+  `reasoning_effort=max` by default; `/model` can switch between V4 Pro and V4
+  Flash with `none`, `high`, or `max` thinking strength.
 - OpenAI Agents SDK integration through `OpenAIChatCompletionsModel`.
 - Project-aware coding tools for reading files, modifying files, running shell
   commands, and showing readable diffs.
@@ -141,15 +142,24 @@ Deepy only uses TOML configuration. JSON config files are intentionally rejected
 api_key = "sk-..."
 name = "deepseek-v4-pro"
 base_url = "https://api.deepseek.com"
+thinking = true
+reasoning_effort = "max" # high or max when thinking is enabled
 
 [context]
 window_tokens = 1048576
 compact_trigger_ratio = 0.8
+reserved_context_tokens = 50000
+compact_preserve_recent_messages = 2
 
 [ui]
 theme = "auto" # auto, dark, or light
 ```
 
+Supported interactive model choices are `deepseek-v4-pro` and
+`deepseek-v4-flash`. In `/model`, thinking strength `none` saves
+`thinking = false`; `high` and `max` save `thinking = true` with the matching
+`reasoning_effort`.
+
 WebSearch uses Deepy's hosted SearXNG endpoint by default. You can override it
 with your own SearXNG instance:
 
@@ -192,8 +202,10 @@ Inside the interactive terminal:
 
 ```text
 /skills   List available skills
+/model    Select model and thinking strength
 /new      Start a fresh conversation
 /resume   Pick a previous session
+/compact  Compact the active session context
 /theme    Show or change UI theme
 /reset    Delete config and run setup again
 /         Open the command menu
@@ -226,6 +238,5 @@ assets live outside the package directory and are not included in the wheel.
 
 ## Release Status
 
-Deepy is preparing its first public `0.1.5` release. The current release path is
-GitHub + PyPI. Standalone binaries and npm wrappers can be added later, but the
-primary distribution is the Python CLI.
+Deepy `0.1.7` is released through GitHub and PyPI. Standalone binaries and npm
+wrappers can be added later, but the primary distribution is the Python CLI.

{deepy_cli-0.1.5 → deepy_cli-0.1.7}/README.md

@@ -34,7 +34,8 @@ context state visible while the agent works.
 ## Highlights
 
 - DeepSeek-first model setup with `deepseek-v4-pro`, thinking enabled, and
-  `reasoning_effort=max` by default.
+  `reasoning_effort=max` by default; `/model` can switch between V4 Pro and V4
+  Flash with `none`, `high`, or `max` thinking strength.
 - OpenAI Agents SDK integration through `OpenAIChatCompletionsModel`.
 - Project-aware coding tools for reading files, modifying files, running shell
   commands, and showing readable diffs.
@@ -113,15 +114,24 @@ Deepy only uses TOML configuration. JSON config files are intentionally rejected
 api_key = "sk-..."
 name = "deepseek-v4-pro"
 base_url = "https://api.deepseek.com"
+thinking = true
+reasoning_effort = "max" # high or max when thinking is enabled
 
 [context]
 window_tokens = 1048576
 compact_trigger_ratio = 0.8
+reserved_context_tokens = 50000
+compact_preserve_recent_messages = 2
 
 [ui]
 theme = "auto" # auto, dark, or light
 ```
 
+Supported interactive model choices are `deepseek-v4-pro` and
+`deepseek-v4-flash`. In `/model`, thinking strength `none` saves
+`thinking = false`; `high` and `max` save `thinking = true` with the matching
+`reasoning_effort`.
+
 WebSearch uses Deepy's hosted SearXNG endpoint by default. You can override it
 with your own SearXNG instance:
 
@@ -164,8 +174,10 @@ Inside the interactive terminal:
 
 ```text
 /skills   List available skills
+/model    Select model and thinking strength
 /new      Start a fresh conversation
 /resume   Pick a previous session
+/compact  Compact the active session context
 /theme    Show or change UI theme
 /reset    Delete config and run setup again
 /         Open the command menu
@@ -198,6 +210,5 @@ assets live outside the package directory and are not included in the wheel.
 
 ## Release Status
 
-Deepy is preparing its first public `0.1.5` release. The current release path is
-GitHub + PyPI. Standalone binaries and npm wrappers can be added later, but the
-primary distribution is the Python CLI.
+Deepy `0.1.7` is released through GitHub and PyPI. Standalone binaries and npm
+wrappers can be added later, but the primary distribution is the Python CLI.

{deepy_cli-0.1.5 → deepy_cli-0.1.7}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "deepy-cli"
-version = "0.1.5"
+version = "0.1.7"
 description = "Deepy - Vibe coding for DeepSeek models in your terminal"
 readme = "README.md"
 authors = [

{deepy_cli-0.1.5 → deepy_cli-0.1.7}/src/deepy/__init__.py

@@ -1,6 +1,6 @@
 from __future__ import annotations
 
-__version__ = "0.1.5"
+__version__ = "0.1.7"
 
 
 def main() -> None:

{deepy_cli-0.1.5 → deepy_cli-0.1.7}/src/deepy/cli.py

@@ -226,6 +226,11 @@ def _doctor(args: argparse.Namespace) -> tuple[int, dict[str, object]]:
         or settings.context.resolved_compact_threshold > 0,
         str(settings.context.resolved_compact_threshold),
     )
+    check(
+        "reserved_context",
+        settings.context.reserved_context_tokens > 0,
+        str(settings.context.reserved_context_tokens),
+    )
 
     try:
         build_provider_bundle(settings)
@@ -241,6 +246,7 @@ def _doctor(args: argparse.Namespace) -> tuple[int, dict[str, object]]:
         "thinking": {
             "enabled": settings.model.thinking_enabled,
             "reasoning_effort": settings.model.reasoning_effort,
+            "reasoning_mode": settings.model.reasoning_mode,
         },
     }
 
@@ -306,7 +312,7 @@ def _cmd_doctor(args: argparse.Namespace) -> int:
         status = "ok" if item["ok"] else "fail"
         print(f"{status:4} {item['name']}: {item['detail']}")
     thinking = report["thinking"]
-    print(f"info thinking: enabled={thinking['enabled']} effort={thinking['reasoning_effort']}")
+    print(f"info reasoning: mode={thinking['reasoning_mode']}")
     live = report.get("live")
     if isinstance(live, dict):
         if live.get("ok"):

{deepy_cli-0.1.5 → deepy_cli-0.1.7}/src/deepy/config/__init__.py

@@ -2,18 +2,27 @@ from __future__ import annotations
 
 from .settings import (
     ContextConfig,
+    DEEPSEEK_MODEL_CATALOG,
+    DEFAULT_COMPACT_PRESERVE_RECENT_MESSAGES,
+    DEFAULT_RESERVED_CONTEXT_TOKENS,
     DEFAULT_UI_THEME,
     DEFAULT_WEB_SEARCH_SEARXNG_URL,
+    DeepSeekModelInfo,
     ModelConfig,
+    REASONING_MODES,
     Settings,
+    SUPPORTED_DEEPSEEK_MODELS,
     UI_THEME_OPTIONS,
     UI_THEMES,
     UiConfig,
     default_config_path,
+    is_supported_deepseek_model,
     is_valid_ui_theme,
+    is_valid_reasoning_mode,
     load_settings,
     mask_secret,
     settings_to_toml_dict,
+    update_config_model_settings,
     update_config_theme,
     ui_theme_from_selection,
     ui_theme_number,
@@ -22,18 +31,27 @@ from .settings import (
 
 __all__ = [
     "ContextConfig",
+    "DEEPSEEK_MODEL_CATALOG",
+    "DEFAULT_COMPACT_PRESERVE_RECENT_MESSAGES",
+    "DEFAULT_RESERVED_CONTEXT_TOKENS",
     "DEFAULT_UI_THEME",
     "DEFAULT_WEB_SEARCH_SEARXNG_URL",
+    "DeepSeekModelInfo",
     "ModelConfig",
+    "REASONING_MODES",
     "Settings",
+    "SUPPORTED_DEEPSEEK_MODELS",
     "UI_THEME_OPTIONS",
     "UI_THEMES",
     "UiConfig",
     "default_config_path",
+    "is_supported_deepseek_model",
     "is_valid_ui_theme",
+    "is_valid_reasoning_mode",
     "load_settings",
     "mask_secret",
     "settings_to_toml_dict",
+    "update_config_model_settings",
     "update_config_theme",
     "ui_theme_from_selection",
     "ui_theme_number",

{deepy_cli-0.1.5 → deepy_cli-0.1.7}/src/deepy/config/settings.py

@@ -12,14 +12,40 @@ DEFAULT_MODEL = "deepseek-v4-pro"
 DEFAULT_BASE_URL = "https://api.deepseek.com"
 DEFAULT_CONTEXT_WINDOW_TOKENS = 1_048_576
 DEFAULT_COMPACT_TRIGGER_RATIO = 0.8
-DEFAULT_COMPACT_PROMPT_TOKEN_THRESHOLD = 838_861
+DEFAULT_RESERVED_CONTEXT_TOKENS = 50_000
+DEFAULT_COMPACT_PRESERVE_RECENT_MESSAGES = 2
 DEFAULT_WEB_SEARCH_SEARXNG_URL = "https://s.kirineko.tech/"
 DEFAULT_UI_THEME = "auto"
 REASONING_EFFORTS = {"high", "max"}
+REASONING_MODES = {"none", "high", "max"}
 UI_THEMES = {"auto", "dark", "light"}
 UI_THEME_OPTIONS = (("1", "auto"), ("2", "dark"), ("3", "light"))
 
 
+@dataclass(frozen=True)
+class DeepSeekModelInfo:
+    name: str
+    label: str
+    description: str
+    supports_thinking: bool = True
+    default_reasoning_mode: str = "max"
+
+
+DEEPSEEK_MODEL_CATALOG = (
+    DeepSeekModelInfo(
+        name="deepseek-v4-pro",
+        label="DeepSeek V4 Pro",
+        description="Higher quality for agentic coding and complex reasoning.",
+    ),
+    DeepSeekModelInfo(
+        name="deepseek-v4-flash",
+        label="DeepSeek V4 Flash",
+        description="Lower latency and cost for faster everyday turns.",
+    ),
+)
+SUPPORTED_DEEPSEEK_MODELS = frozenset(model.name for model in DEEPSEEK_MODEL_CATALOG)
+
+
 def default_config_path() -> Path:
     return Path.home() / ".deepy" / "config.toml"
 
@@ -96,14 +122,22 @@ class ModelConfig:
     def thinking_enabled(self) -> bool:
         if self.thinking is not None:
             return self.thinking
-        return self.name.lower() in {"deepseek-v4-pro", "deepseek-v4-flash"}
+        return self.name.lower() in SUPPORTED_DEEPSEEK_MODELS
+
+    @property
+    def reasoning_mode(self) -> str:
+        if not self.thinking_enabled:
+            return "none"
+        return self.reasoning_effort if self.reasoning_effort in REASONING_EFFORTS else "max"
 
 
 @dataclass(frozen=True)
 class ContextConfig:
     window_tokens: int = DEFAULT_CONTEXT_WINDOW_TOKENS
     compact_trigger_ratio: float = DEFAULT_COMPACT_TRIGGER_RATIO
-    compact_prompt_token_threshold: int | None = None
+    reserved_context_tokens: int = DEFAULT_RESERVED_CONTEXT_TOKENS
+    compact_preserve_recent_messages: int = DEFAULT_COMPACT_PRESERVE_RECENT_MESSAGES
+    compact_preserve_recent_tokens: int | None = None
 
     @classmethod
     def from_mapping(cls, raw: Mapping[str, Any]) -> Self:
@@ -111,18 +145,30 @@ class ContextConfig:
         ratio = _as_float(raw.get("compact_trigger_ratio"), DEFAULT_COMPACT_TRIGGER_RATIO)
         if ratio <= 0 or ratio > 1:
             ratio = DEFAULT_COMPACT_TRIGGER_RATIO
-        threshold = raw.get("compact_prompt_token_threshold")
-        compact_threshold = _as_int(threshold, 0) if threshold is not None else None
+        reserved_context_tokens = _as_int(
+            raw.get("reserved_context_tokens"),
+            DEFAULT_RESERVED_CONTEXT_TOKENS,
+        )
+        preserve_recent_messages = _as_int(
+            raw.get("compact_preserve_recent_messages"),
+            DEFAULT_COMPACT_PRESERVE_RECENT_MESSAGES,
+        )
+        preserve_recent_tokens_raw = raw.get("compact_preserve_recent_tokens")
+        preserve_recent_tokens = (
+            _as_int(preserve_recent_tokens_raw, 0)
+            if preserve_recent_tokens_raw is not None
+            else None
+        )
         return cls(
             window_tokens=window_tokens,
             compact_trigger_ratio=ratio,
-            compact_prompt_token_threshold=compact_threshold or None,
+            reserved_context_tokens=reserved_context_tokens,
+            compact_preserve_recent_messages=preserve_recent_messages,
+            compact_preserve_recent_tokens=preserve_recent_tokens or None,
         )
 
     @property
     def resolved_compact_threshold(self) -> int:
-        if self.compact_prompt_token_threshold:
-            return self.compact_prompt_token_threshold
         return int(self.window_tokens * self.compact_trigger_ratio + 0.999999)
 
 
@@ -234,9 +280,6 @@ def settings_to_toml_dict(settings: Settings, *, reveal_secret: bool = False) ->
     if api_key:
         data["model"]["api_key"] = api_key if reveal_secret else mask_secret(api_key)
     data["model"]["thinking"] = settings.model.thinking_enabled
-    data["context"]["compact_prompt_token_threshold"] = (
-        settings.context.resolved_compact_threshold
-    )
     return _drop_empty(data)
 
 
@@ -244,6 +287,14 @@ def is_valid_ui_theme(value: str) -> bool:
     return value in UI_THEMES
 
 
+def is_supported_deepseek_model(value: str) -> bool:
+    return value in SUPPORTED_DEEPSEEK_MODELS
+
+
+def is_valid_reasoning_mode(value: str) -> bool:
+    return value in REASONING_MODES
+
+
 def ui_theme_number(theme: str) -> str:
     for number, value in UI_THEME_OPTIONS:
         if value == theme:
@@ -288,7 +339,8 @@ def write_config(
         "context": {
             "window_tokens": DEFAULT_CONTEXT_WINDOW_TOKENS,
             "compact_trigger_ratio": DEFAULT_COMPACT_TRIGGER_RATIO,
-            "compact_prompt_token_threshold": DEFAULT_COMPACT_PROMPT_TOKEN_THRESHOLD,
+            "reserved_context_tokens": DEFAULT_RESERVED_CONTEXT_TOKENS,
+            "compact_preserve_recent_messages": DEFAULT_COMPACT_PRESERVE_RECENT_MESSAGES,
         },
         "logging": {
             "debug": False,
@@ -311,23 +363,59 @@ def write_config(
     os.chmod(path, 0o600)
 
 
+def update_config_model_settings(
+    config_path: Path,
+    *,
+    model: str | None = None,
+    reasoning_mode: str | None = None,
+) -> None:
+    if model is not None and not is_supported_deepseek_model(model):
+        raise ValueError(
+            "Model must be one of: " + ", ".join(model_info.name for model_info in DEEPSEEK_MODEL_CATALOG)
+        )
+    if reasoning_mode is not None and not is_valid_reasoning_mode(reasoning_mode):
+        raise ValueError("Reasoning mode must be one of: none, high, max.")
+    path = config_path.expanduser()
+    if path.suffix == ".json":
+        raise ValueError("Deepy only supports TOML config files; JSON config is not supported.")
+    raw = _read_toml_mapping(path)
+    model_section = raw.get("model")
+    model_map = dict(model_section) if isinstance(model_section, Mapping) else {}
+    if model is not None:
+        model_map["name"] = model
+    if reasoning_mode is not None:
+        if reasoning_mode == "none":
+            model_map["thinking"] = False
+        else:
+            model_map["thinking"] = True
+            model_map["reasoning_effort"] = reasoning_mode
+    raw["model"] = model_map
+    _write_private_toml(path, raw)
+
+
 def update_config_theme(config_path: Path, theme: str) -> None:
     if not is_valid_ui_theme(theme):
         raise ValueError("UI theme must be one of: auto, dark, light.")
     path = config_path.expanduser()
     if path.suffix == ".json":
         raise ValueError("Deepy only supports TOML config files; JSON config is not supported.")
-    raw: dict[str, Any]
-    if path.exists():
-        with path.open("rb") as fh:
-            loaded = tomllib.load(fh)
-        raw = dict(loaded)
-    else:
-        raw = {}
+    raw = _read_toml_mapping(path)
     ui = raw.get("ui")
     ui_map = dict(ui) if isinstance(ui, Mapping) else {}
     ui_map["theme"] = theme
     raw["ui"] = ui_map
+    _write_private_toml(path, raw)
+
+
+def _read_toml_mapping(path: Path) -> dict[str, Any]:
+    if not path.exists():
+        return {}
+    with path.open("rb") as fh:
+        loaded = tomllib.load(fh)
+    return dict(loaded)
+
+
+def _write_private_toml(path: Path, raw: Mapping[str, Any]) -> None:
     path.parent.mkdir(parents=True, exist_ok=True)
     path.write_text(tomli_w.dumps(raw), encoding="utf-8")
     os.chmod(path, 0o600)

{deepy_cli-0.1.5 → deepy_cli-0.1.7}/src/deepy/data/tools/AskUserQuestion.md

@@ -1,6 +1,8 @@
 ## AskUserQuestion
 
-Ask the user only when progress is blocked by missing intent or a required decision.
+Ask the user when clarification would materially improve the result: ambiguous intent,
+unclear scope, user preferences, high-impact trade-offs, or required approval. Do not
+ask for low-impact details when a reasonable assumption can keep progress moving.
 
 Args: `questions` (non-empty array). Each question needs `question` and non-empty `options`;
 each option needs `label` and may include `description`. Use `multiSelect=true` only when

{deepy_cli-0.1.5 → deepy_cli-0.1.7}/src/deepy/data/tools/WebFetch.md

@@ -5,5 +5,6 @@ Fetch a specific web page when the user provides a complete URL.
 Args: `url`.
 
 Accepts only complete `http://` or `https://` URLs. Returns the final URL, title,
-content type, and extracted readable text for HTML pages. Use `WebSearch` to
+content type, and extracted readable text for HTML pages, including standard
+description metadata when ordinary body text is unavailable. Use `WebSearch` to
 discover URLs; use `WebFetch` when the URL is already known.

deepy_cli-0.1.7/src/deepy/data/tools/shell.md

@@ -0,0 +1,13 @@
+## shell
+
+Run commands in the detected runtime shell for inspection, tests, builds, and
+project operations.
+
+Args: `command`, optional `timeout_ms`.
+
+Use the runtime context's command dialect and path style: PowerShell uses
+PowerShell commands and Windows paths, `cmd` uses cmd syntax, and `posix` uses
+POSIX shell syntax.
+
+Runs in the session cwd, preserves cwd between calls when supported, and returns
+stdout/stderr JSON with cwd, exit-code, and shell metadata.

{deepy_cli-0.1.5 → deepy_cli-0.1.7}/src/deepy/llm/__init__.py

@@ -1,6 +1,6 @@
 from __future__ import annotations
 
-from .context import build_session_input_callback
+from .context import build_session_input_callback, should_auto_compact
 from .events import DeepyStreamEvent, normalize_stream_event
 from .thinking import build_model_settings, build_thinking_extra_body
 
@@ -10,4 +10,5 @@ __all__ = [
     "build_session_input_callback",
     "build_thinking_extra_body",
     "normalize_stream_event",
+    "should_auto_compact",
 ]