Skip to content

armgit / camel

Public Imported from GitHub
0

🐫 CAMEL: The first and the best multi-agent framework. Finding the Scaling Law of Agents. https://www.camel-ai.org

7.1 KB
Raw Blame
<a id="camel.toolkits.agent_toolkit"></a>

<a id="camel.toolkits.agent_toolkit.AgentToolkit"></a>

## AgentToolkit

```python
class AgentToolkit(BaseToolkit, RegisteredAgentToolkit):
```

Toolkit for delegating a task to a persistent specialized sub-agent.

This toolkit exposes tools for either waiting on a delegated task directly
or starting one and checking its result later.

<a id="camel.toolkits.agent_toolkit.AgentToolkit.__init__"></a>

### __init__

```python
def __init__(self, timeout: Optional[float] = None):
```

Initialize the AgentToolkit.

**Parameters:**

- **timeout** (Optional[float]): Maximum execution time for toolkit calls. (default: :obj:`None`)

<a id="camel.toolkits.agent_toolkit.AgentToolkit._build_system_message"></a>

### _build_system_message

```python
def _build_system_message(self, subagent_type: str, description: str):
```

<a id="camel.toolkits.agent_toolkit.AgentToolkit._error_result"></a>

### _error_result

```python
def _error_result(self, error_message: str, **payload: Any):
```

<a id="camel.toolkits.agent_toolkit.AgentToolkit._require_parent_agent"></a>

### _require_parent_agent

```python
def _require_parent_agent(self):
```

<a id="camel.toolkits.agent_toolkit.AgentToolkit._resolve_child_tools"></a>

### _resolve_child_tools

```python
def _resolve_child_tools(self, parent: Optional['ChatAgent']):
```

<a id="camel.toolkits.agent_toolkit.AgentToolkit._create_subagent"></a>

### _create_subagent

```python
def _create_subagent(self, subagent_type: str, description: str):
```

<a id="camel.toolkits.agent_toolkit.AgentToolkit._run_agent_step"></a>

### _run_agent_step

```python
def _run_agent_step(self, agent: 'ChatAgent', prompt: str):
```

<a id="camel.toolkits.agent_toolkit.AgentToolkit._complete_task"></a>

### _complete_task

```python
def _complete_task(self, task_id: str):
```

<a id="camel.toolkits.agent_toolkit.AgentToolkit._submit_agent_task"></a>

### _submit_agent_task

```python
def _submit_agent_task(
    self,
    agent_id: str,
    agent: 'ChatAgent',
    prompt: str
):
```

Submit a new task. Caller MUST hold `__INLINE_CODE_0__`.

<a id="camel.toolkits.agent_toolkit.AgentToolkit._get_task"></a>

### _get_task

```python
def _get_task(self, task_id: str):
```

<a id="camel.toolkits.agent_toolkit.AgentToolkit.agent_run_subagent"></a>

### agent_run_subagent

```python
def agent_run_subagent(
    self,
    prompt: str,
    description: str = 'Specialized sub-agent task',
    subagent_type: str = 'general-purpose',
    agent_id: Optional[str] = None,
    wait: bool = True,
    timeout: Optional[float] = None
):
```

Run a specialized sub-agent, optionally waiting for completion.

Use this tool when the current task is complex enough to benefit from
a focused child agent with its own conversation state. If
:obj:`agent_id` is omitted, a new sub-agent session is created with a
fresh ChatAgent instance and its own empty conversation state. If
:obj:`agent_id` is provided, the matching sub-agent instance is reused
rather than recreated. This means the sub-agent keeps its existing
memory and any state accumulated from previous turns in that session.

Prefer :obj:`wait=True` when the parent agent needs the sub-agent
result before it can continue. Use :obj:`wait=False` only when the
parent agent intentionally wants to do other work first and check the
task later with :obj:`agent_get_task_output`.

**Parameters:**

- **prompt** (str): The task instructions to send to the sub-agent for this turn.
- **description** (str): Short description of the sub-task or role for the sub-agent. This is primarily used when creating a new session. (default: :obj:`"Specialized sub-agent task"`)
- **subagent_type** (str): Specialization label for the spawned sub-agent, such as :obj:`general-purpose`, :obj:`research`, :obj:`analysis`, :obj:`coding`, or :obj:`writing`. (default: :obj:`"general-purpose"`)
- **agent_id** (Optional[str]): Existing sub-agent session ID to resume. When provided, the stored sub-agent instance is reused, so it continues with the same memory / conversation state from earlier turns. In this case, :obj:`description` and :obj:`subagent_type` are ignored. The spawned sub-agent always uses the calling parent agent's model. (default: :obj:`None`)
- **wait** (bool): Whether to wait for the sub-agent to finish before returning. Set to :obj:`False` to start the task and check it later with :obj:`agent_get_task_output`. (default: :obj:`True`)
- **timeout** (Optional[float]): Maximum wait time in seconds for the sub-agent to finish when :obj:`wait` is True. If the timeout is reached, the task keeps running and the current status is returned.

**Returns:**

  Dict[str, Any]: Final result for completed tasks, or current task
metadata if still running.

<a id="camel.toolkits.agent_toolkit.AgentToolkit.agent_get_task_output"></a>

### agent_get_task_output

```python
def agent_get_task_output(
    self,
    task_id: str,
    block: bool = False,
    timeout: Optional[float] = None
):
```

Get the latest result of a sub-agent task started earlier.

Use this after calling :obj:`agent_run_subagent(wait=False)`. When
:obj:`block` is False, the call returns immediately with the latest
known status. When :obj:`block` is True, the tool waits up to
:obj:`timeout` seconds for completion before returning.

**Parameters:**

- **task_id** (str): Sub-agent task ID returned by :obj:`agent_run_subagent`.
- **block** (bool): Whether to wait for the task to finish before returning status and output. (default: :obj:`False`)
- **timeout** (Optional[float]): Maximum wait time in seconds when :obj:`block` is True. (default: :obj:`None`)

**Returns:**

  Dict[str, Any]: Task status, associated :obj:`agent_id`, and any
available result or error output.

<a id="camel.toolkits.agent_toolkit.AgentToolkit.agent_stop_task"></a>

### agent_stop_task

```python
def agent_stop_task(self, task_id: str):
```

Request cancellation of a running sub-agent task.

**Parameters:**

- **task_id** (str): Sub-agent task ID returned by :obj:`agent_run_subagent`.

**Returns:**

  Dict[str, Any]: Stop request status and the latest known task
state.

<a id="camel.toolkits.agent_toolkit.AgentToolkit.clone_for_new_session"></a>

### clone_for_new_session

```python
def clone_for_new_session(self, new_session_id: Optional[str] = None):
```

Create a fresh AgentToolkit without carrying over sub-agent state.

**Parameters:**

- **new_session_id** (Optional[str]): Unused compatibility parameter for ChatAgent toolkit cloning. (default: :obj:`None`)

**Returns:**

  AgentToolkit: A new toolkit instance with the same configuration
and no active sub-agent sessions.

<a id="camel.toolkits.agent_toolkit.AgentToolkit._purge_completed_tasks"></a>

### _purge_completed_tasks

```python
def _purge_completed_tasks(self):
```

Remove finished tasks to prevent unbounded memory growth.

<a id="camel.toolkits.agent_toolkit.AgentToolkit.cleanup"></a>

### cleanup

```python
def cleanup(self):
```

<a id="camel.toolkits.agent_toolkit.AgentToolkit.__del__"></a>

### __del__

```python
def __del__(self):
```

<a id="camel.toolkits.agent_toolkit.AgentToolkit.get_tools"></a>

### get_tools

```python
def get_tools(self):
```

Return the tool list for this toolkit.