yaicli 0.0.8__tar.gz → 0.0.11__tar.gz

{yaicli-0.0.8 → yaicli-0.0.11}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: yaicli
-Version: 0.0.8
+Version: 0.0.11
 Summary: A simple CLI tool to interact with LLM
 Project-URL: Homepage, https://github.com/belingud/yaicli
 Project-URL: Repository, https://github.com/belingud/yaicli
@@ -214,21 +214,24 @@ Classifier: Operating System :: OS Independent
 Classifier: Programming Language :: Python :: 3
 Requires-Python: >=3.9
 Requires-Dist: distro>=1.9.0
+Requires-Dist: httpx>=0.28.1
 Requires-Dist: jmespath>=1.0.1
 Requires-Dist: prompt-toolkit>=3.0.50
-Requires-Dist: requests>=2.32.3
 Requires-Dist: rich>=13.9.4
+Requires-Dist: socksio>=1.0.0
 Requires-Dist: typer>=0.15.2
 Description-Content-Type: text/markdown
 
-# YAICLI - Your AI Command Line Interface
+# YAICLI - Your AI Interface in Command Line
 
 [![PyPI version](https://img.shields.io/pypi/v/yaicli?style=for-the-badge)](https://pypi.org/project/yaicli/)
 ![GitHub License](https://img.shields.io/github/license/belingud/yaicli?style=for-the-badge)
 ![PyPI - Downloads](https://img.shields.io/pypi/dm/yaicli?logo=pypi&style=for-the-badge)
 ![Pepy Total Downloads](https://img.shields.io/pepy/dt/yaicli?style=for-the-badge&logo=python)
 
-YAICLI is a powerful command-line AI assistant tool that enables you to interact with Large Language Models (LLMs) like ChatGPT's gpt-4o through your terminal. It offers multiple operation modes for everyday conversations, generating and executing shell commands, and one-shot quick queries.
+YAICLI is a compact yet potent command-line AI assistant, allowing you to engage with Large Language Models (LLMs) such as ChatGPT's gpt-4o directly via your terminal. It offers multiple operation modes for everyday conversations, generating and executing shell commands, and one-shot quick queries.
+
+Support regular and deep thinking models.
 
 > [!WARNING] 
 > This is a work in progress, some features could change or be removed in the future.
@@ -257,6 +260,9 @@ YAICLI is a powerful command-line AI assistant tool that enables you to interact
 - **Keyboard Shortcuts**:
   - Tab to switch between Chat and Execute modes
 
+- **History**:
+  - Save and recall previous queries
+
 ## Installation
 
 ### Prerequisites
@@ -296,6 +302,7 @@ The default configuration file is located at `~/.config/yaicli/config.ini`. Look
 
 ```ini
 [core]
+PROVIDER=OPENAI
 BASE_URL=https://api.openai.com/v1
 API_KEY=your_api_key_here
 MODEL=gpt-4o
@@ -312,6 +319,10 @@ ANSWER_PATH=choices[0].message.content
 # true: streaming response
 # false: non-streaming response
 STREAM=true
+
+TEMPERATURE=0.7
+TOP_P=1.0
+MAX_TOKENS=1024
 ```
 
 ### Configuration Options
@@ -327,6 +338,58 @@ Below are the available configuration options and override environment variables
 - **ANSWER_PATH**: Json path expression to extract answer from response, default: choices[0].message.content, env: AI_ANSWER_PATH
 - **STREAM**: Enable/disable streaming responses, default: true, env: AI_STREAM
 
+Default config of `COMPLETION_PATH` and `ANSWER_PATH` is OpenAI compatible. If you are using OpenAI or other OpenAI compatible LLM provider, you can use the default config.
+
+If you wish to use other providers that are not compatible with the openai interface, you can use the following config:
+
+- claude:
+  - BASE_URL: https://api.anthropic.com/v1
+  - COMPLETION_PATH: /messages
+  - ANSWER_PATH: content.0.text
+- cohere:
+  - BASE_URL: https://api.cohere.com/v2
+  - COMPLETION_PATH: /chat
+  - ANSWER_PATH: message.content.[0].text
+- google:
+  - BASE_URL: https://generativelanguage.googleapis.com/v1beta/openai
+  - COMPLETION_PATH: /chat/completions
+  - ANSWER_PATH: choices[0].message.content
+
+You can use google OpenAI complete endpoint and leave `COMPLETION_PATH` and `ANSWER_PATH` as default. BASE_URL: https://generativelanguage.googleapis.com/v1beta/openai. See https://ai.google.dev/gemini-api/docs/openai
+
+Claude also has a testable OpenAI-compatible interface, you can just use Calude endpoint and leave `COMPLETION_PATH` and `ANSWER_PATH` as default. See: https://docs.anthropic.com/en/api/openai-sdk
+
+If you not sure how to config `COMPLETION_PATH` and `ANSWER_PATH`, here is a guide:
+1. **Find the API Endpoint**:
+   - Visit the documentation of the LLM provider you want to use.
+   - Find the API endpoint for the completion task. This is usually under the "API Reference" or "Developer Documentation" section.
+2. **Identify the Response Structure**:
+   - Look for the structure of the response. This typically includes fields like `choices`, `completion`, etc.
+3. **Identify the Path Expression**:
+   Forexample, claude response structure like this:
+   ```json
+      {
+      "content": [
+        {
+          "text": "Hi! My name is Claude.",
+          "type": "text"
+        }
+      ],
+      "id": "msg_013Zva2CMHLNnXjNJJKqJ2EF",
+      "model": "claude-3-7-sonnet-20250219",
+      "role": "assistant",
+      "stop_reason": "end_turn",
+      "stop_sequence": null,
+      "type": "message",
+      "usage": {
+        "input_tokens": 2095,
+        "output_tokens": 503
+      }
+    }
+   ```
+    We are looking for the `text` field, so the path should be 1.Key `content`, 2.First obj `[0]`, 3.Key `text`. So it should be `content.[0].text`.
+
+
 ## Usage
 
 ### Basic Usage
@@ -456,4 +519,4 @@ Contributions of code, issue reports, or feature suggestions are welcome.
 
 ---
 
-*YAICLI - Making your terminal smarter*
+*YAICLI - Making your terminal smarter*

{yaicli-0.0.8 → yaicli-0.0.11}/README.md

@@ -1,11 +1,13 @@
-# YAICLI - Your AI Command Line Interface
+# YAICLI - Your AI Interface in Command Line
 
 [![PyPI version](https://img.shields.io/pypi/v/yaicli?style=for-the-badge)](https://pypi.org/project/yaicli/)
 ![GitHub License](https://img.shields.io/github/license/belingud/yaicli?style=for-the-badge)
 ![PyPI - Downloads](https://img.shields.io/pypi/dm/yaicli?logo=pypi&style=for-the-badge)
 ![Pepy Total Downloads](https://img.shields.io/pepy/dt/yaicli?style=for-the-badge&logo=python)
 
-YAICLI is a powerful command-line AI assistant tool that enables you to interact with Large Language Models (LLMs) like ChatGPT's gpt-4o through your terminal. It offers multiple operation modes for everyday conversations, generating and executing shell commands, and one-shot quick queries.
+YAICLI is a compact yet potent command-line AI assistant, allowing you to engage with Large Language Models (LLMs) such as ChatGPT's gpt-4o directly via your terminal. It offers multiple operation modes for everyday conversations, generating and executing shell commands, and one-shot quick queries.
+
+Support regular and deep thinking models.
 
 > [!WARNING] 
 > This is a work in progress, some features could change or be removed in the future.
@@ -34,6 +36,9 @@ YAICLI is a powerful command-line AI assistant tool that enables you to interact
 - **Keyboard Shortcuts**:
   - Tab to switch between Chat and Execute modes
 
+- **History**:
+  - Save and recall previous queries
+
 ## Installation
 
 ### Prerequisites
@@ -73,6 +78,7 @@ The default configuration file is located at `~/.config/yaicli/config.ini`. Look
 
 ```ini
 [core]
+PROVIDER=OPENAI
 BASE_URL=https://api.openai.com/v1
 API_KEY=your_api_key_here
 MODEL=gpt-4o
@@ -89,6 +95,10 @@ ANSWER_PATH=choices[0].message.content
 # true: streaming response
 # false: non-streaming response
 STREAM=true
+
+TEMPERATURE=0.7
+TOP_P=1.0
+MAX_TOKENS=1024
 ```
 
 ### Configuration Options
@@ -104,6 +114,58 @@ Below are the available configuration options and override environment variables
 - **ANSWER_PATH**: Json path expression to extract answer from response, default: choices[0].message.content, env: AI_ANSWER_PATH
 - **STREAM**: Enable/disable streaming responses, default: true, env: AI_STREAM
 
+Default config of `COMPLETION_PATH` and `ANSWER_PATH` is OpenAI compatible. If you are using OpenAI or other OpenAI compatible LLM provider, you can use the default config.
+
+If you wish to use other providers that are not compatible with the openai interface, you can use the following config:
+
+- claude:
+  - BASE_URL: https://api.anthropic.com/v1
+  - COMPLETION_PATH: /messages
+  - ANSWER_PATH: content.0.text
+- cohere:
+  - BASE_URL: https://api.cohere.com/v2
+  - COMPLETION_PATH: /chat
+  - ANSWER_PATH: message.content.[0].text
+- google:
+  - BASE_URL: https://generativelanguage.googleapis.com/v1beta/openai
+  - COMPLETION_PATH: /chat/completions
+  - ANSWER_PATH: choices[0].message.content
+
+You can use google OpenAI complete endpoint and leave `COMPLETION_PATH` and `ANSWER_PATH` as default. BASE_URL: https://generativelanguage.googleapis.com/v1beta/openai. See https://ai.google.dev/gemini-api/docs/openai
+
+Claude also has a testable OpenAI-compatible interface, you can just use Calude endpoint and leave `COMPLETION_PATH` and `ANSWER_PATH` as default. See: https://docs.anthropic.com/en/api/openai-sdk
+
+If you not sure how to config `COMPLETION_PATH` and `ANSWER_PATH`, here is a guide:
+1. **Find the API Endpoint**:
+   - Visit the documentation of the LLM provider you want to use.
+   - Find the API endpoint for the completion task. This is usually under the "API Reference" or "Developer Documentation" section.
+2. **Identify the Response Structure**:
+   - Look for the structure of the response. This typically includes fields like `choices`, `completion`, etc.
+3. **Identify the Path Expression**:
+   Forexample, claude response structure like this:
+   ```json
+      {
+      "content": [
+        {
+          "text": "Hi! My name is Claude.",
+          "type": "text"
+        }
+      ],
+      "id": "msg_013Zva2CMHLNnXjNJJKqJ2EF",
+      "model": "claude-3-7-sonnet-20250219",
+      "role": "assistant",
+      "stop_reason": "end_turn",
+      "stop_sequence": null,
+      "type": "message",
+      "usage": {
+        "input_tokens": 2095,
+        "output_tokens": 503
+      }
+    }
+   ```
+    We are looking for the `text` field, so the path should be 1.Key `content`, 2.First obj `[0]`, 3.Key `text`. So it should be `content.[0].text`.
+
+
 ## Usage
 
 ### Basic Usage
@@ -233,4 +295,4 @@ Contributions of code, issue reports, or feature suggestions are welcome.
 
 ---
 
-*YAICLI - Making your terminal smarter*
+*YAICLI - Making your terminal smarter*

{yaicli-0.0.8 → yaicli-0.0.11}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "yaicli"
-version = "0.0.8"
+version = "0.0.11"
 description = "A simple CLI tool to interact with LLM"
 authors = [{ name = "belingud", email = "im.victor@qq.com" }]
 readme = "README.md"
@@ -31,10 +31,11 @@ keywords = [
 ]
 dependencies = [
     "distro>=1.9.0",
+    "httpx>=0.28.1",
     "jmespath>=1.0.1",
     "prompt-toolkit>=3.0.50",
-    "requests>=2.32.3",
     "rich>=13.9.4",
+    "socksio>=1.0.0",
     "typer>=0.15.2",
 ]
 [project.urls]
@@ -51,11 +52,12 @@ resolution = "highest"
 [dependency-groups]
 dev = ["bump2version>=1.0.1", "pytest>=8.3.5", "ruff>=0.11.2"]
 
+[tool.isort]
+profile = "black"
+
 [tool.ruff]
 line-length = 120
-select = ["E", "F", "W", "I", "B", "C90"]
-ignore = ["E501"]
-
+fix = true
 
 [build-system]
 requires = ["hatchling>=1.18.0"]

{yaicli-0.0.8 → yaicli-0.0.11}/yaicli.py

@@ -2,14 +2,13 @@ import configparser
 import json
 import platform
 import subprocess
-import time
 from os import getenv
 from os.path import basename, pathsep
 from pathlib import Path
-from typing import Annotated, Optional
+from typing import Annotated, Optional, Union
 
+import httpx
 import jmespath
-import requests
 import typer
 from distro import name as distro_name
 from prompt_toolkit import PromptSession
@@ -30,7 +29,7 @@ Rules:
 5. Return NOTHING except the ready-to-run command"""
 
 DEFAULT_PROMPT = (
-    "You are yaili, a system management and programing assistant, "
+    "You are YAICLI, a system management and programing assistant, "
     "You are managing {_os} operating system with {_shell} shell. "
     "Your responses should be concise and use Markdown format, "
     "unless the user explicitly requests more details."
@@ -54,23 +53,7 @@ DEFAULT_CONFIG_MAP = {
     "STREAM": {"value": "true", "env_key": "AI_STREAM"},
 }
 
-app = typer.Typer(
-    name="yaicli",
-    context_settings={"help_option_names": ["-h", "--help"]},
-    pretty_exceptions_enable=False,
-)
-
-
-class CasePreservingConfigParser(configparser.RawConfigParser):
-    """Case preserving config parser"""
-
-    def optionxform(self, optionstr):
-        return optionstr
-
-
-class CLI:
-    CONFIG_PATH = Path("~/.config/yaicli/config.ini").expanduser()
-    DEFAULT_CONFIG_INI = """[core]
+DEFAULT_CONFIG_INI = """[core]
 PROVIDER=openai
 BASE_URL=https://api.openai.com/v1
 API_KEY=
@@ -87,7 +70,28 @@ ANSWER_PATH=choices[0].message.content
 
 # true: streaming response
 # false: non-streaming response
-STREAM=true"""
+STREAM=true
+
+TEMPERATURE=0.7
+TOP_P=1.0
+MAX_TOKENS=1024"""
+
+app = typer.Typer(
+    name="yaicli",
+    context_settings={"help_option_names": ["-h", "--help"]},
+    pretty_exceptions_enable=False,
+)
+
+
+class CasePreservingConfigParser(configparser.RawConfigParser):
+    """Case preserving config parser"""
+
+    def optionxform(self, optionstr):
+        return optionstr
+
+
+class CLI:
+    CONFIG_PATH = Path("~/.config/yaicli/config.ini").expanduser()
 
     def __init__(self, verbose: bool = False) -> None:
         self.verbose = verbose
@@ -123,7 +127,7 @@ STREAM=true"""
             self.console.print("[bold yellow]Creating default configuration file.[/bold yellow]")
             self.CONFIG_PATH.parent.mkdir(parents=True, exist_ok=True)
             with open(self.CONFIG_PATH, "w") as f:
-                f.write(self.DEFAULT_CONFIG_INI)
+                f.write(DEFAULT_CONFIG_INI)
         else:
             # Load from configuration file (middle priority)
             config_parser = CasePreservingConfigParser()
@@ -180,10 +184,11 @@ STREAM=true"""
         example:
         ```bash\nls -la\n``` ==> ls -al
         ```zsh\nls -la\n``` ==> ls -al
-        ```ls -al``` ==> ls -al
-        ls -al ==> ls -al
+        ```ls -la``` ==> ls -la
+        ls -la ==> ls -la
         ```\ncd /tmp\nls -la\n``` ==> cd /tmp\nls -la
         ```bash\ncd /tmp\nls -la\n``` ==> cd /tmp\nls -la
+        ```plaintext\nls -la\n``` ==> ls -la
         """
         if not command or not command.strip():
             return ""
@@ -211,65 +216,130 @@ STREAM=true"""
             # Join the remaining lines and strip any extra whitespace
             return "\n".join(line.strip() for line in content_lines if line.strip())
 
-    def post(self, message: list[dict[str, str]]) -> requests.Response:
+    def _get_type_number(self, key, _type: type, default=None):
+        """Get number with type from config"""
+        try:
+            return _type(self.config.get(key, default))
+        except ValueError:
+            raise ValueError(f"[red]{key} should be {_type} type.[/red]")
+
+    def post(self, message: list[dict[str, str]]) -> httpx.Response:
         """Post message to LLM API and return response"""
         url = self.config.get("BASE_URL", "").rstrip("/") + "/" + self.config.get("COMPLETION_PATH", "").lstrip("/")
         body = {
             "messages": message,
             "model": self.config.get("MODEL", "gpt-4o"),
             "stream": self.config.get("STREAM", "true") == "true",
-            "temperature": 0.7,
-            "top_p": 1,
+            "temperature": self._get_type_number(key="TEMPERATURE", _type=float, default="0.7"),
+            "top_p": self._get_type_number(key="TOP_P", _type=float, default="1.0"),
+            "max_tokens": self._get_type_number(key="MAX_TOKENS", _type=int, default="1024"),
         }
-        response = requests.post(url, json=body, headers={"Authorization": f"Bearer {self.config.get('API_KEY', '')}"})
+        with httpx.Client(timeout=120.0) as client:
+            response = client.post(
+                url, json=body, headers={"Authorization": f"Bearer {self.config.get('API_KEY', '')}"}
+            )
         try:
             response.raise_for_status()
-        except requests.exceptions.HTTPError as e:
+        except httpx.HTTPStatusError as e:
             self.console.print(f"[red]Error calling API: {e}[/red]")
             if self.verbose:
-                self.console.print(f"Reason: {e.response.reason}")
+                self.console.print(f"Reason: {e}")
                 self.console.print(f"Response: {response.text}")
-            raise typer.Exit(code=1) from None
+            raise e
         return response
 
-    def _print(self, response: requests.Response, stream: bool = True) -> str:
-        """Print response from LLM and return full completion"""
+    def get_reasoning_content(self, delta: dict) -> Optional[str]:
+        # reasoning: openrouter
+        # reasoning_content: infi-ai/deepseek
+        for k in ("reasoning_content", "reasoning"):
+            if k in delta:
+                return delta[k]
+        return None
+
+    def _parse_stream_line(self, line: Union[bytes, str]) -> Optional[dict]:
+        """Parse a single line from the stream response"""
+        if not line:
+            return None
+
+        if isinstance(line, bytes):
+            line = line.decode("utf-8")
+        if not line.startswith("data: "):
+            return None
+
+        line = line[6:]
+        if line == "[DONE]":
+            return None
+
+        try:
+            json_data = json.loads(line)
+            if not json_data.get("choices"):
+                return None
+
+            return json_data
+        except json.JSONDecodeError:
+            self.console.print("[red]Error decoding response JSON[/red]")
+            if self.verbose:
+                self.console.print(f"[red]Error JSON data: {line}[/red]")
+            return None
+
+    def _process_reasoning_content(self, reason: str, full_completion: str, in_reasoning: bool) -> tuple[str, bool]:
+        """Process reasoning content in the response"""
+        if not in_reasoning:
+            in_reasoning = True
+            full_completion = "> Reasoning:\n> "
+        full_completion += reason.replace("\n", "\n> ")
+        return full_completion, in_reasoning
+
+    def _process_regular_content(self, content: str, full_completion: str, in_reasoning: bool) -> tuple[str, bool]:
+        """Process regular content in the response"""
+        if in_reasoning:
+            in_reasoning = False
+            full_completion += "\n\n"
+        full_completion += content
+        return full_completion, in_reasoning
+
+    def _print_stream(self, response: httpx.Response) -> str:
+        """Print response from LLM in streaming mode"""
         full_completion = ""
+        in_reasoning = False
+
+        with Live() as live:
+            for line in response.iter_lines():
+                json_data = self._parse_stream_line(line)
+                if not json_data:
+                    continue
+
+                delta = json_data["choices"][0]["delta"]
+                reason = self.get_reasoning_content(delta)
+
+                if reason is not None:
+                    full_completion, in_reasoning = self._process_reasoning_content(
+                        reason, full_completion, in_reasoning
+                    )
+                else:
+                    content = delta.get("content", "") or ""
+                    full_completion, in_reasoning = self._process_regular_content(
+                        content, full_completion, in_reasoning
+                    )
+
+                live.update(Markdown(markup=full_completion), refresh=True)
+
+        return full_completion
+
+    def _print_non_stream(self, response: httpx.Response) -> str:
+        """Print response from LLM in non-streaming mode"""
+        full_completion = jmespath.search(self.config.get("ANSWER_PATH", "choices[0].message.content"), response.json())
+        self.console.print(Markdown(full_completion))
+        return full_completion
+
+    def _print(self, response: httpx.Response, stream: bool = True) -> str:
+        """Print response from LLM and return full completion"""
         if stream:
-            with Live() as live:
-                for line in response.iter_lines():
-                    # Skip empty lines
-                    if not line:
-                        continue
-
-                    # Process server-sent events
-                    data = line.decode("utf-8")
-                    if not data.startswith("data: "):
-                        continue
-
-                    # Extract data portion
-                    data = data[6:]
-                    if data == "[DONE]":
-                        break
-
-                    # Parse JSON and update display
-                    try:
-                        json_data = json.loads(data)
-                        content = json_data["choices"][0]["delta"].get("content", "")
-                        full_completion += content
-                        live.update(Markdown(markup=full_completion), refresh=True)
-                    except json.JSONDecodeError:
-                        self.console.print("[red]Error decoding response JSON[/red]")
-                        if self.verbose:
-                            self.console.print(f"[red]Error: {data}[/red]")
-
-                    time.sleep(0.01)
+            # Streaming response
+            full_completion = self._print_stream(response)
         else:
             # Non-streaming response
-            full_completion = jmespath.search(
-                self.config.get("ANSWER_PATH", "choices[0].message.content"), response.json()
-            )
-            self.console.print(Markdown(full_completion))
+            full_completion = self._print_non_stream(response)
         self.console.print()  # Add a newline after the response to separate from the next input
         return full_completion
 
@@ -292,7 +362,13 @@ STREAM=true"""
         """Run REPL loop, handling user input and generating responses, saving history, and executing commands"""
         # Show REPL instructions
         self._setup_key_bindings()
-        self.console.print("[bold]Starting REPL loop[/bold]")
+        self.console.print("""
+██    ██  █████  ██  ██████ ██      ██
+ ██  ██  ██   ██ ██ ██      ██      ██
+  ████   ███████ ██ ██      ██      ██
+   ██    ██   ██ ██ ██      ██      ██
+   ██    ██   ██ ██  ██████ ███████ ██
+""")
         self.console.print("[bold]Press TAB to change in chat and exec mode[/bold]")
         self.console.print("[bold]Type /clear to clear chat history[/bold]")
         self.console.print("[bold]Type /his to see chat history[/bold]")
@@ -328,12 +404,21 @@ STREAM=true"""
             message.append({"role": "user", "content": user_input})
 
             # Get response from LLM
-            response = self.post(message)
+            try:
+                response = self.post(message)
+            except ValueError as e:
+                self.console.print(f"[red]Error: {e}[/red]")
+                return
+            except httpx.ConnectError as e:
+                self.console.print(f"[red]Error: {e}[/red]")
+                continue
+            except httpx.HTTPStatusError:
+                continue
             self.console.print("\n[bold green]Assistant:[/bold green]")
             try:
                 content = self._print(response, stream=self.config["STREAM"] == "true")
             except Exception as e:
-                self.console.print(f"[red]Error: {e}[/red]")
+                self.console.print(f"[red]Unknown Error: {e}[/red]")
                 continue
 
             # Add user input and assistant response to history
@@ -355,24 +440,10 @@ STREAM=true"""
 
         self.console.print("[bold green]Exiting...[/bold green]")
 
-    def run(self, chat: bool, shell: bool, prompt: str) -> None:
-        """Run the CLI"""
-        self.load_config()
-        if not self.config.get("API_KEY"):
-            self.console.print("[bold red]API key not set[/bold red]")
-            self.console.print(
-                "[bold red]Please set API key in ~/.config/yaicli/config.ini or environment variable[/bold red]"
-            )
-            raise typer.Exit(code=1)
+    def _run_once(self, prompt: str, shell: bool = False) -> None:
+        """Run once with given prompt"""
         _os = self.detect_os()
         _shell = self.detect_shell()
-
-        # Handle chat mode
-        if chat:
-            self.current_mode = CHAT_MODE
-            self._run_repl()
-            return
-
         # Create appropriate system prompt based on mode
         system_prompt = SHELL_PROMPT if shell else DEFAULT_PROMPT
         system_content = system_prompt.format(_os=_os, _shell=_shell)
@@ -384,7 +455,14 @@ STREAM=true"""
         ]
 
         # Get response from LLM
-        response = self.post(message)
+        try:
+            response = self.post(message)
+        except (ValueError, httpx.ConnectError, httpx.HTTPStatusError) as e:
+            self.console.print(f"[red]Error: {e}[/red]")
+            return
+        except Exception as e:
+            self.console.print(f"[red]Unknown Error: {e}[/red]")
+            return
         self.console.print("\n[bold green]Assistant:[/bold green]")
         content = self._print(response, stream=self.config["STREAM"] == "true")
 
@@ -400,6 +478,23 @@ STREAM=true"""
                 if returncode != 0:
                     self.console.print(f"[bold red]Command failed with return code {returncode}[/bold red]")
 
+    def run(self, chat: bool, shell: bool, prompt: str) -> None:
+        """Run the CLI"""
+        self.load_config()
+        if not self.config.get("API_KEY"):
+            self.console.print("[bold red]API key not set[/bold red]")
+            self.console.print(
+                "[bold red]Please set API key in ~/.config/yaicli/config.ini or environment variable[/bold red]"
+            )
+            raise typer.Exit(code=1)
+
+        # Handle chat mode
+        if chat:
+            self.current_mode = CHAT_MODE
+            self._run_repl()
+        else:
+            self._run_once(prompt, shell)
+
 
 @app.command()
 def main(