Skip to content

extending.md

Latest commit

 

History

History
111 lines (78 loc) · 3.3 KB

extending.md

File metadata and controls

111 lines (78 loc) · 3.3 KB

Extending Pipe

This guide explains how to extend the pipe framework by adding new agents and integrations.

Adding New Agents

To add support for a new AI agent (e.g., Claude API, Claude Code, or other LLM providers), follow these steps:

1. Create Agent Implementation

Create a new Python file in src/pipe/core/agents/ for your agent. Use existing agents like gemini_api.py or gemini_cli.py as references.

Example structure for claude_api.py:

import requests
from typing import Dict, Any
from src.pipe.core.models import AgentResponse

class ClaudeApiAgent:
    def __init__(self, api_key: str):
        self.api_key = api_key
        self.base_url = "https://api.anthropic.com/v1/messages"

    def execute(self, prompt: str, **kwargs) -> AgentResponse:
        headers = {
            "x-api-key": self.api_key,
            "Content-Type": "application/json",
            "anthropic-version": "2023-06-01"
        }

        data = {
            "model": "claude-3-sonnet-20240229",
            "max_tokens": 4096,
            "messages": [{"role": "user", "content": prompt}]
        }

        response = requests.post(self.base_url, headers=headers, json=data)
        response.raise_for_status()

        result = response.json()
        return AgentResponse(
            content=result["content"][0]["text"],
            metadata={"model": "claude-3-sonnet", "usage": result.get("usage", {})}
        )

2. Update Dispatcher

Add your new agent to src/pipe/core/dispatcher.py. Import your agent class and add it to the agent mapping.

Example:

from src.pipe.core.agents.claude_api import ClaudeApiAgent

# In the dispatcher class
def get_agent(self, agent_type: str):
    agents = {
        "gemini_api": GeminiApiAgent,
        "gemini_cli": GeminiCliAgent,
        "claude_api": ClaudeApiAgent,  # Add your new agent
    }
    return agents[agent_type](**self.config)

3. Create Prompt Template

Create a new Jinja2 template in templates/prompt/ based on gemini_api_prompt.j2. Adapt it for your agent's specific requirements.

Example claude_api_prompt.j2:

You are Claude, an AI assistant built by Anthropic.

{% include 'roles.j2' %}
{% include 'description.j2' %}
{% include 'constraints.j2' %}
{% include 'conversation_history.j2' %}
{% include 'current_task.j2' %}

Please respond as Claude would, following the guidelines above.

4. Update Configuration

Add support for your new agent in the configuration system (setting.yml or environment variables). Ensure API keys and other settings are properly handled.

5. Add Tests

Create unit tests for your new agent in the tests/ directory, following the existing test patterns.

Best Practices

  • Follow the existing code style and patterns
  • Handle errors gracefully and provide meaningful error messages
  • Include proper logging for debugging
  • Document your agent implementation
  • Test with various prompt sizes and edge cases

Example: Adding Claude Code Integration

For integrating with Claude Code (similar to Gemini CLI):

  1. Create src/pipe/core/agents/claude_code.py
  2. Implement CLI interaction logic
  3. Add to dispatcher as "claude_code"
  4. Create appropriate prompt template
  5. Update configuration options

For more information on the codebase structure, see docs/development.md.