Skip to content

agent

Hooks to modify the Cat's Agent.

Here is a collection of methods to hook into the Agent execution pipeline.

agent_allowed_tools(allowed_tools, cat)

Hook the allowed tools.

Allows to decide which tools end up in the Agent prompt.

To decide, you can filter the list of tools' names, but you can also check the context in cat.working_memory and launch custom chains with cat._llm.

Parameters:

Name Type Description Default
cat CheshireCat

Cheshire Cat instance.

 required

Returns:

Name Type Description
tools List[str]

List of allowed Langchain tools.
Source code in cat/mad_hatter/core_plugin/hooks/agent.py 
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
@hook(priority=0)
def agent_allowed_tools(allowed_tools: List[str], cat) -> List[str]:
    """Hook the allowed tools.

    Allows to decide which tools end up in the *Agent* prompt.

    To decide, you can filter the list of tools' names, but you can also check the context in `cat.working_memory`
    and launch custom chains with `cat._llm`.

    Parameters
    ---------
    cat : CheshireCat
        Cheshire Cat instance.

    Returns
    -------
    tools : List[str]
        List of allowed Langchain tools.
    """

    return allowed_tools

agent_fast_reply(agent_fast_reply, cat)

This hook allows for a custom response after memory recall, skipping default agent execution. It's useful for custom agent logic or when you want to use recalled memories but avoid the main agent.

Parameters:

Name Type Description Default
agent_fast_reply dict

An initially empty dict that can be populated with a response.

 required
cat CheshireCat

Cheshire Cat instance.

 required

Returns:

Name Type Description
response None | dict | AgentOutput

If you want to bypass the main agent, return an AgentOutput or a dict with an "output" key. Return None to continue with normal execution. See below for examples of Cat response

Examples:

Example 1: don't remember (no uploaded documents about topic) 

num_declarative_memories = len( cat.working_memory.declarative_memories )
if num_declarative_memories == 0:
    return {
       "output": "Sorry, I have no memories about that."
    }
Source code in cat/mad_hatter/core_plugin/hooks/agent.py 
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
@hook(priority=0)
def agent_fast_reply(agent_fast_reply: dict, cat) -> None | dict | AgentOutput:
    """This hook allows for a custom response after memory recall, skipping default agent execution.
    It's useful for custom agent logic or when you want to use recalled memories but avoid the main agent.

    Parameters
    --------
    agent_fast_reply: dict
        An initially empty dict that can be populated with a response.
    cat : CheshireCat
        Cheshire Cat instance.

    Returns
    --------
    response : None | dict | AgentOutput
        If you want to bypass the main agent, return an AgentOutput or a dict with an "output" key.
        Return None to continue with normal execution.
        See below for examples of Cat response

    Examples
    --------

    Example 1: don't remember (no uploaded documents about topic)
    ```python
    num_declarative_memories = len( cat.working_memory.declarative_memories )
    if num_declarative_memories == 0:
        return {
           "output": "Sorry, I have no memories about that."
        }
    ```
    """

    return agent_fast_reply

before_agent_starts(agent_input, cat)

Hook to read and edit the agent input

Parameters:

Name Type Description Default
agent_input Dict

Input that is about to be passed to the agent.

 required
cat CheshireCat

Cheshire Cat instance.

 required

Returns:

Name Type Description
response Dict

Agent Input
Source code in cat/mad_hatter/core_plugin/hooks/agent.py 
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
@hook(priority=0)
def before_agent_starts(agent_input: Dict, cat) -> Dict:
    """Hook to read and edit the agent input

    Parameters
    --------
    agent_input: dict
        Input that is about to be passed to the agent.
    cat : CheshireCat
        Cheshire Cat instance.

    Returns
    --------
    response : Dict
        Agent Input
    """

    return agent_input