# IBMi Langgraph Agents
## Available Agents
- **Performance Agent**: System performance monitoring and analysis
- **SysAdmin Discovery Agent**: High-level system discovery and summarization
- **SysAdmin Browse Agent**: Detailed system browsing and exploration
- **SysAdmin Search Agent**: System search and lookup capabilities
## Prerequisites
- IBM i MCP Server running at http://127.0.0.1:3010/mcp (or configured URL)
- Python 3.13 or higher
- uv package manager
## Setup
### 1. Ensure the IBM i MCP Server is Running
The MCP server should be running at the URL specified in your `.env` file (default: http://127.0.0.1:3010/mcp).
You can check if the server is running with:
```bash
ps aux | grep ibmi-mcp-server
```
### 2. Configure Environment Variables
Create or update the `.env` file in the `agents/frameworks/langchain` directory
### 3. Install Dependencies
First, install the ibmi-agent-sdk package in editable mode:
```bash
cd /path/to/ibmi-mcp-server/agents/packages/ibmi-agent-sdk
uv pip install -e .
```
Then, install the LangChain agents package:
```bash
cd /path/to/ibmi-mcp-server/agents/frameworks/langchain
uv pip install -e .
```
## Usage
### Quick Test
To verify that all agents can be created successfully:
```bash
cd /path/to/ibmi-mcp-server/agents/frameworks/langchain/src/ibmi_agents/agents
uv run test_agents.py --quick
```
### Test Specific Agent
To test a specific agent with a sample query:
```bash
uv run test_agents.py --agent performance
```
Available agent types: `performance`, `discovery`, `browse`, `search`
### Interactive Mode
To chat interactively with a specific agent:
```bash
uv run test_agents.py --agent performance --interactive
```
### Additional Options
- `--model`: Specify a different model (e.g., `--model openai:gpt-4o`)
- `--quiet`: Disable verbose logging (only show final responses)
- `--verbose`: Enable verbose logging (default)
## Using Agents in Your Code
You can import and use the agents directly in your Python code:
```python
import asyncio
from ibmi_agents import create_agent, chat_with_agent
async def main():
# Create a performance agent
ctx = await create_agent("performance", model_id="ollama:llama3.2")
# Use the agent in an async context
async with ctx as (agent, session):
# Send a query to the agent
response = await chat_with_agent(
agent,
"What is my system status?",
thread_id="my-session-1"
)
print(response)
if __name__ == "__main__":
asyncio.run(main())
```
### Available Functions in ibmi_agents.py
- `create_agent(agent_type, **kwargs)`: Create an agent of the specified type
- `chat_with_agent(agent, message, thread_id, verbose)`: Send a message to an agent and get the response
- `list_available_agents()`: Get information about all available agent types
- `set_verbose_logging(enabled)`: Enable or disable verbose logging globally
- `get_verbose_logging()`: Get current verbose logging status
### Agent Types
- `performance`: System performance monitoring and analysis
- `discovery`: High-level system discovery and summarization
- `browse`: Detailed system browsing and exploration
- `search`: System search and lookup capabilities
### Model Configuration
You can specify different models when creating agents:
```python
# Using Ollama (local models)
ctx = await create_agent("performance", model_id="ollama:llama3.2")
# Using OpenAI (requires API key)
ctx = await create_agent("performance", model_id="openai:gpt-4o")
# Using Anthropic (requires API key)
ctx = await create_agent("performance", model_id="anthropic:claude-3.7-sonnet")
```
