CLI Attachments

Overview

CLI providers (geminicli:, claudecli:, codex:, vibe:, opencode:) support file and folder attachments through the extra_body.cli parameter. This enables the AI to read, analyze, and modify local files.

CLI attachments are only supported with CLI providers. API providers (gemini:, claude:, etc.) do not support this feature.

Syntax

Add attachments via the extra_body.cli.attachments array:

{
  "model": "geminicli:gemini-2.5-pro",
  "messages": [{"role": "user", "content": "Analyze this code"}],
  "extra_body": {
    "cli": {
      "attachments": [
        {"type": "file", "path": "/path/to/file.py"},
        {"type": "folder", "path": "./src/"}
      ]
    }
  }
}

Attachment Types

File Attachments

Attach a single file:

{
  "type": "file",
  "path": "/absolute/or/relative/path/to/file.py"
}

Supported Path Formats:

Absolute: /home/user/project/main.py
Relative: ./src/main.py (relative to server working directory)
Home directory: ~/projects/file.py

Folder Attachments

Attach an entire directory:

{
  "type": "folder",
  "path": "./src/"
}

All files in the folder and subdirectories are included.

Examples

Single File

curl http://localhost:18080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer sk-test-123" \
  -d '{
    "model": "geminicli:gemini-2.5-pro",
    "messages": [
      {"role": "user", "content": "Fix the bugs in this code"}
    ],
    "extra_body": {
      "cli": {
        "attachments": [
          {"type": "file", "path": "./main.py"}
        ]
      }
    }
  }'

from openai import OpenAI

client = OpenAI(
    base_url="http://localhost:18080/v1",
    api_key="sk-test-123"
)

response = client.chat.completions.create(
    model="geminicli:gemini-2.5-pro",
    messages=[
        {"role": "user", "content": "Fix the bugs in this code"}
    ],
    extra_body={
        "cli": {
            "attachments": [
                {"type": "file", "path": "./main.py"}
            ]
        }
    }
)

print(response.choices[0].message.content)

import OpenAI from 'openai';

const client = new OpenAI({
  baseURL: 'http://localhost:18080/v1',
  apiKey: 'sk-test-123'
});

const response = await client.chat.completions.create({
  model: 'geminicli:gemini-2.5-pro',
  messages: [
    { role: 'user', content: 'Fix the bugs in this code' }
  ],
  extra_body: {
    cli: {
      attachments: [
        { type: 'file', path: './main.py' }
      ]
    }
  }
});

console.log(response.choices[0].message.content);

Multiple Files

response = client.chat.completions.create(
    model="geminicli:gemini-2.5-pro",
    messages=[
        {"role": "user", "content": "Review this pull request"}
    ],
    extra_body={
        "cli": {
            "attachments": [
                {"type": "file", "path": "./src/main.py"},
                {"type": "file", "path": "./src/utils.py"},
                {"type": "file", "path": "./tests/test_main.py"}
            ]
        }
    }
)

Folder Attachment

response = client.chat.completions.create(
    model="claudecli:claude-sonnet-4",
    messages=[
        {"role": "user", "content": "Refactor this codebase to use TypeScript"}
    ],
    extra_body={
        "cli": {
            "attachments": [
                {"type": "folder", "path": "./src/"}
            ]
        }
    }
)

Mixed Attachments

Combine files and folders:

response = client.chat.completions.create(
    model="geminicli:gemini-2.5-pro",
    messages=[
        {"role": "user", "content": "Add tests for the new features"}
    ],
    extra_body={
        "cli": {
            "attachments": [
                {"type": "folder", "path": "./src/"},       # Source code
                {"type": "folder", "path": "./tests/"},     # Existing tests
                {"type": "file", "path": "./package.json"}  # Dependencies
            ]
        }
    }
)

CLI Flags

Control CLI behavior with flags:

{
  "extra_body": {
    "cli": {
      "attachments": [...],
      "flags": {
        "sandbox": true,
        "auto_approve": true,
        "yolo": false
      }
    }
  }
}

Available Flags

boolean

default:false

Run in restricted execution mode. Prevents file modifications.CLI Mapping:

geminicli: -s or --sandbox
codex: --sandbox
claudecli: Not supported
vibe: Not supported

boolean

default:false

Skip confirmation prompts for file operations.CLI Mapping:

geminicli: -y or --yes
vibe: --auto-approve
claudecli: --dangerously-skip-permissions
codex: --full-auto

boolean

default:false

Maximum autonomy mode. Combines auto-approve with other permissive flags.CLI Mapping:

geminicli: --yolo
vibe: --auto-approve
claudecli: --dangerously-skip-permissions
codex: --full-auto

Flag Examples

Sandbox Mode
Auto-Approve
YOLO Mode

# Read-only mode - AI can view but not modify files
response = client.chat.completions.create(
    model="geminicli:gemini-2.5-pro",
    messages=[
        {"role": "user", "content": "Analyze code quality"}
    ],
    extra_body={
        "cli": {
            "attachments": [{"type": "folder", "path": "./src/"}],
            "flags": {
                "sandbox": True  # Read-only
            }
        }
    }
)

# Skip confirmations for file modifications
response = client.chat.completions.create(
    model="claudecli:claude-sonnet-4",
    messages=[
        {"role": "user", "content": "Fix all linting errors"}
    ],
    extra_body={
        "cli": {
            "attachments": [{"type": "folder", "path": "./src/"}],
            "flags": {
                "auto_approve": True  # Skip prompts
            }
        }
    }
)

# Maximum autonomy - use with caution!
response = client.chat.completions.create(
    model="geminicli:gemini-2.5-pro",
    messages=[
        {"role": "user", "content": "Implement the feature"}
    ],
    extra_body={
        "cli": {
            "attachments": [{"type": "folder", "path": "./"}],
            "flags": {
                "yolo": True  # Full autonomy
            }
        }
    }
)

YOLO mode grants maximum autonomy. Only use in controlled environments or with proper backups.

Session Management

Resume or name CLI sessions:

response = client.chat.completions.create(
    model="geminicli:gemini-2.5-pro",
    messages=[
        {"role": "user", "content": "Continue working on the feature"}
    ],
    extra_body={
        "cli": {
            "session_id": "my-project-session",
            "attachments": [{"type": "folder", "path": "./"}]
        }
    }
)

Session ID Values

Value	Behavior
`"latest"`	Resume most recent session
`"new"`	Force new session
`"my-session-name"`	Named session (resumable)

Session Example

# First request: Start named session
response1 = client.chat.completions.create(
    model="geminicli:gemini-2.5-pro",
    messages=[
        {"role": "user", "content": "Create a user authentication system"}
    ],
    extra_body={
        "cli": {
            "session_id": "auth-feature",
            "attachments": [{"type": "folder", "path": "./src/"}]
        }
    }
)

# Later request: Resume same session
response2 = client.chat.completions.create(
    model="geminicli:gemini-2.5-pro",
    messages=[
        {"role": "user", "content": "Add password reset functionality"}
    ],
    extra_body={
        "cli": {
            "session_id": "auth-feature",  # Same session
            "attachments": [{"type": "folder", "path": "./src/"}]
        }
    }
)

Use Cases

Code Review

response = client.chat.completions.create(
    model="geminicli:gemini-2.5-pro",
    messages=[
        {
            "role": "user",
            "content": "Review this code for security vulnerabilities and suggest improvements"
        }
    ],
    extra_body={
        "cli": {
            "attachments": [
                {"type": "folder", "path": "./src/"},
                {"type": "file", "path": "./package.json"}
            ],
            "flags": {"sandbox": True}  # Read-only review
        }
    }
)

Automated Refactoring

response = client.chat.completions.create(
    model="claudecli:claude-sonnet-4",
    messages=[
        {
            "role": "user",
            "content": "Refactor to use async/await throughout"
        }
    ],
    extra_body={
        "cli": {
            "attachments": [{"type": "folder", "path": "./src/"}],
            "flags": {"auto_approve": True}  # Auto-apply changes
        }
    }
)

Test Generation

response = client.chat.completions.create(
    model="geminicli:gemini-2.5-pro",
    messages=[
        {
            "role": "user",
            "content": "Generate comprehensive unit tests for all functions"
        }
    ],
    extra_body={
        "cli": {
            "attachments": [
                {"type": "folder", "path": "./src/"},
                {"type": "folder", "path": "./tests/"}  # Existing tests for reference
            ]
        }
    }
)

Documentation Generation

response = client.chat.completions.create(
    model="codex:gpt-5",
    messages=[
        {
            "role": "user",
            "content": "Add comprehensive JSDoc comments to all functions"
        }
    ],
    extra_body={
        "cli": {
            "attachments": [{"type": "folder", "path": "./src/"}],
            "flags": {"auto_approve": True}
        }
    }
)

Codebase Migration

response = client.chat.completions.create(
    model="geminicli:gemini-2.5-pro",
    messages=[
        {
            "role": "user",
            "content": "Migrate from JavaScript to TypeScript. Add type definitions to all files."
        }
    ],
    extra_body={
        "cli": {
            "attachments": [
                {"type": "folder", "path": "./src/"},
                {"type": "file", "path": "./tsconfig.json"}
            ],
            "session_id": "ts-migration",
            "flags": {"auto_approve": True}
        }
    }
)

Provider Support

Provider	Files	Folders	Flags	Sessions
geminicli	✅	✅	✅	✅
claudecli	✅	✅	⚠️	❌
codex	✅	✅	✅	✅
vibe	✅	✅	⚠️	❌
opencode	✅	✅	✅	✅

⚠️ = Partial support (limited flags)

Security Considerations

Path Validation

switchAILocal validates all file paths to prevent directory traversal attacks:

# Safe: Relative paths are resolved securely
{"type": "file", "path": "./src/main.py"}

# Unsafe: Absolute paths outside workspace are rejected
{"type": "file", "path": "/etc/passwd"}  # Blocked

Sandbox Mode

Use sandbox mode for untrusted operations:

# Safe: Read-only analysis
extra_body={
    "cli": {
        "attachments": [...],
        "flags": {"sandbox": True}
    }
}

File Size Limits

Large files may exceed CLI limits:

config.yaml

cli:
  max_attachment_size: 10485760  # 10MB
  max_total_size: 52428800       # 50MB

Sensitive Files

Avoid attaching sensitive files:

# Don't attach:
# - .env files
# - credentials.json
# - private keys
# - API tokens

Error Handling

from openai import OpenAI, APIError

client = OpenAI(
    base_url="http://localhost:18080/v1",
    api_key="sk-test-123"
)

try:
    response = client.chat.completions.create(
        model="geminicli:gemini-2.5-pro",
        messages=[
            {"role": "user", "content": "Analyze this file"}
        ],
        extra_body={
            "cli": {
                "attachments": [
                    {"type": "file", "path": "./nonexistent.py"}
                ]
            }
        }
    )
except APIError as e:
    if "file not found" in str(e).lower():
        print("File does not exist")
    elif "permission denied" in str(e).lower():
        print("Cannot access file")
    else:
        print(f"Error: {e.message}")

import OpenAI from 'openai';

const client = new OpenAI({
  baseURL: 'http://localhost:18080/v1',
  apiKey: 'sk-test-123'
});

try {
  const response = await client.chat.completions.create({
    model: 'geminicli:gemini-2.5-pro',
    messages: [
      { role: 'user', content: 'Analyze this file' }
    ],
    extra_body: {
      cli: {
        attachments: [
          { type: 'file', path: './nonexistent.py' }
        ]
      }
    }
  });
} catch (error) {
  if (error.message.includes('file not found')) {
    console.log('File does not exist');
  } else if (error.message.includes('permission denied')) {
    console.log('Cannot access file');
  } else {
    console.error('Error:', error.message);
  }
}

Limitations

Limitation	Value	Notes
Max file size	10 MB	Configurable in `config.yaml`
Max total size	50 MB	Sum of all attachments
Max files	100	Per request
Supported formats	All text	Binary files may fail

Best Practices

Start with Sandbox: Test with read-only mode first
Use Relative Paths: More portable and secure
Minimize Attachments: Only include relevant files
Name Sessions: Use descriptive session IDs for continuity
Validate Outputs: Review AI changes before committing

Next Steps

Provider Prefixes

Learn about CLI provider routing

Chat Completions

Master the chat completions API

Configuration

Configure CLI provider settings

Overview

Endpoints

Provider Formats

Management API

Overview

Syntax

Attachment Types

File Attachments

Folder Attachments

Examples

Single File

Multiple Files

Folder Attachment

Mixed Attachments

CLI Flags

Available Flags

Flag Examples

Session Management

Session ID Values

Session Example

Use Cases

Code Review

Automated Refactoring

Test Generation

Documentation Generation

Codebase Migration

Provider Support

Security Considerations

Error Handling

Limitations

Best Practices

Next Steps

Provider Prefixes

Chat Completions

Configuration

​Overview

​Syntax

​Attachment Types

​File Attachments

​Folder Attachments

​Examples

​Single File

​Multiple Files

​Folder Attachment

​Mixed Attachments

​CLI Flags

​Available Flags

​Flag Examples

​Session Management

​Session ID Values

​Session Example

​Use Cases

​Code Review

​Automated Refactoring

​Test Generation

​Documentation Generation

​Codebase Migration

​Provider Support

​Security Considerations

​Error Handling

​Limitations

​Best Practices

​Next Steps

Provider Prefixes

Chat Completions

Configuration

Overview

Syntax

Attachment Types

File Attachments

Folder Attachments

Examples

Single File

Multiple Files

Folder Attachment

Mixed Attachments

CLI Flags

Available Flags

Flag Examples

Session Management

Session ID Values

Session Example

Use Cases

Code Review

Automated Refactoring

Test Generation

Documentation Generation

Codebase Migration

Provider Support

Security Considerations

Error Handling

Limitations

Best Practices

Next Steps