# multillm-cli

Command-line interface for [multillm](../multillm).

## Installation

```bash
pip install multillm-cli
```

You'll also need to install provider packages:

```bash
pip install multillm-openai multillm-anthropic multillm-claude
```

## Usage

```bash
multillm -m <model> -p <prompt>
```

### Examples

**Chat providers:**

```bash
# OpenAI
multillm -m openai/gpt-4o -p "What is 2+2?"

# Anthropic
multillm -m anthropic/claude-sonnet-4-20250514 -p "Explain async/await in Python"

# Gemini
multillm -m gemini/gemini-2.0-flash-exp -p "What are the benefits of Rust?"
```

**Piping input from stdin:**

```bash
# Summarize a file
cat document.txt | multillm -m openai/gpt-4o -p "Summarize this:" --with-stdin

# Analyze code
cat script.py | multillm -m claude/default -p "Review this code:" --with-stdin

# Process command output
ls -la | multillm -m anthropic/claude-sonnet-4-20250514 -p "Explain these files:" --with-stdin
```

**Chat providers with built-in tools:**

```bash
# Calculate
multillm -m openai/gpt-4o -p "What is 15 * 23?" --use-tools calculate

# Get current time
multillm -m openai/gpt-4o -p "What time is it?" --use-tools get_current_time

# Weather (mock data)
multillm -m openai/gpt-4o -p "What's the weather in Tokyo?" --use-tools get_weather

# Multiple tools
multillm -m openai/gpt-4o \
  -p "What's 5+5 and what time is it?" \
  --use-tools calculate get_current_time

# Verbose mode (shows tool arguments and results)
multillm -m openai/gpt-4o \
  -p "Calculate 100 / 4" \
  --use-tools calculate \
  --verbose
```

**Agent provider with tools:**

```bash
# Simple query
multillm -m claude/default -p "What is 2+2?"

# With tool use
multillm -m claude/default \
  -p "What Python version is installed?" \
  --allowed-tools Bash \
  --permission-mode acceptEdits

# Complex task with verbose output
multillm -m claude/default \
  -p "List all .py files in the current directory" \
  --allowed-tools Bash Read \
  --permission-mode acceptEdits \
  --max-turns 10 \
  --verbose
```

## Options

| Option | Description |
|--------|-------------|
| `-m`, `--model` | Model to use (format: `provider/model-name`) |
| `-p`, `--prompt` | Prompt to send to the model |
| `--with-stdin` | Append stdin to the prompt after a separator |
| `--use-tools` | Enable built-in tools for chat providers (see below) |
| `--max-turns` | Maximum turns for agent providers |
| `--allowed-tools` | Space-separated list of allowed tools for agent providers |
| `--permission-mode` | Permission mode: `acceptEdits`, `acceptAll`, or `prompt` |
| `-v`, `--verbose` | Show detailed tool execution information |

## Built-in Tools (Chat Providers)

When using chat providers (OpenAI, Anthropic, Gemini, OpenRouter), you can enable built-in tools with `--use-tools`:

| Tool | Description | Example |
|------|-------------|---------|
| `calculate` | Perform mathematical calculations | `--use-tools calculate` |
| `get_current_time` | Get current date and time | `--use-tools get_current_time` |
| `get_weather` | Get weather information (mock data) | `--use-tools get_weather` |
| `ask_user` | Ask the user a question interactively | `--use-tools ask_user` |

### Interactive Tools

The `ask_user` tool allows the model to ask you questions during execution and collect your responses. This enables truly interactive conversations where the model can clarify requirements, gather preferences, or get additional information.

**Example:**
```bash
multillm -m openai/gpt-4o \
  -p "Help me choose a programming language for my project by asking about my requirements" \
  --use-tools ask_user
```

When the model calls `ask_user`, you'll see:
```
======================================================================
❓ QUESTION FROM ASSISTANT
======================================================================

What type of project are you building?

Suggested options:
  1. Web application
  2. Desktop application
  3. Data science
  4. Command-line tool

You can select a number or provide your own answer.

Your answer: _
```

### Tool Output

When tools are used, you'll see inline output showing:

**Normal mode:**
```
[Using 1 tool(s)]
  → calculate()
  ✓ Result: 345

The result of 15 * 23 is 345.
```

**Verbose mode (`--verbose`):**
```
[Using 1 tool(s)]
  → calculate({"expression": "15 * 23"})
  ← {
    "expression": "15 * 23",
    "result": 345
  }

The result of 15 * 23 is 345.
```

### Example Usage

```bash
# Basic calculation
multillm -m openai/gpt-4o -p "What is 123 + 456?" --use-tools calculate

# With verbose output
multillm -m openai/gpt-4o -p "Calculate 50 / 2" --use-tools calculate -v

# Multiple tools
multillm -m openai/gpt-4o \
  -p "What's the current time and what's 10 * 10?" \
  --use-tools get_current_time calculate

# All tools enabled
multillm -m openai/gpt-4o \
  -p "What time is it in Tokyo and what's the weather like?" \
  --use-tools get_current_time get_weather \
  --verbose
```

## Available Tools (Agent Providers)

When using agent providers (like `claude`), you can specify which tools the agent can use with `--allowed-tools`:

### Core Tools

| Tool | Description |
|------|-------------|
| `Read` | Read files from the filesystem |
| `Write` | Create or overwrite files |
| `Edit` | Edit existing files with find/replace |
| `Bash` | Execute bash commands |
| `Glob` | Find files using glob patterns |
| `Grep` | Search file contents with regex |

### Extended Tools

| Tool | Description |
|------|-------------|
| `Task` | Launch specialized sub-agents for complex tasks |
| `WebFetch` | Fetch and process web content |
| `WebSearch` | Search the web for information |
| `NotebookEdit` | Edit Jupyter notebook cells |
| `AskUserQuestion` | Ask the user questions during execution |
| `KillShell` | Kill running background shells |
| `EnterPlanMode` | Enter planning mode for complex implementations |
| `ExitPlanMode` | Exit planning mode |

### Usage Examples

```bash
# Basic file operations
multillm -m claude/default -p "Read README.md" --allowed-tools Read

# Command execution
multillm -m claude/default -p "What Python version?" --allowed-tools Bash

# Multiple tools
multillm -m claude/default -p "Find all .py files and read them" \
  --allowed-tools Glob Read

# Common combination for development tasks
multillm -m claude/default -p "Fix the bug in main.py" \
  --allowed-tools Read Write Edit Bash Grep
```

**Note:** The more tools you allow, the more autonomous the agent becomes. Start with minimal tools and add more as needed.

## Configuration

Provider credentials can be set via:

**Option 1: Config files** (recommended)

Create provider configs at `~/.config/multillm/providers/<provider>.json`.

Example for OpenAI (`~/.config/multillm/providers/openai.json`):
```json
{
  "api_key": "sk-..."
}
```

Set permissions:
```bash
chmod 600 ~/.config/multillm/providers/*.json
```

See individual provider READMEs for all configuration options:
- [multillm-openai](../multillm-openai)
- [multillm-anthropic](../multillm-anthropic)
- [multillm-gemini](../multillm-gemini)
- [multillm-openrouter](../multillm-openrouter)
- [multillm-claude](../multillm-claude)

**Option 2: Environment variables**

```bash
export OPENAI_API_KEY=sk-...
export ANTHROPIC_API_KEY=sk-ant-...
export GOOGLE_API_KEY=...
export OPENROUTER_API_KEY=sk-or-...
```

**For Claude agent provider:**

```bash
# OAuth via CLI (recommended)
claude login

# Or API key
export ANTHROPIC_API_KEY=sk-ant-...

# Or OAuth token
export CLAUDE_CODE_OAUTH_TOKEN=your-token
```

## Supported Providers

- **Chat providers**: `openai`, `anthropic`, `gemini`, `openrouter`
- **Agent providers**: `claude`