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 
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
@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(fast_reply, cat)

This hook is useful to shortcut the Cat response. If you do not want the agent to run, return the final response from here and it will end up in the chat without the agent being executed.

Parameters:

Name Type Description Default
fast_reply

Input is dict (initially empty), which can be enriched whith an "output" key with the shortcut response.

 required
cat CheshireCat

Cheshire Cat instance.

 required

Returns:

Name Type Description
response Union[None, Dict]

Cat response if you want to avoid using the agent, or None / {} if you want the agent to be executed. See below for examples of Cat response

Examples:

Example 1: can't talk about this topic 

# here you could use cat._llm to do topic evaluation
if "dog" in agent_input["input"]:
    return {
        "output": "You went out of topic. Can't talk about dog."
    }

Example 2: 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 
32
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
66
67
68
69
70
71
72
@hook(priority=0)
def agent_fast_reply(fast_reply, cat) -> Union[None, Dict]:
    """This hook is useful to shortcut the Cat response.
    If you do not want the agent to run, return the final response from here and it will end up in the chat without the agent being executed.

    Parameters
    --------
    fast_reply: dict
        Input is dict (initially empty), which can be enriched whith an "output" key with the shortcut response.
    cat : CheshireCat
        Cheshire Cat instance.

    Returns
    --------
    response : Union[None, Dict]
        Cat response if you want to avoid using the agent, or None / {} if you want the agent to be executed.
        See below for examples of Cat response

    Examples
    --------

    Example 1: can't talk about this topic
    ```python
    # here you could use cat._llm to do topic evaluation
    if "dog" in agent_input["input"]:
        return {
            "output": "You went out of topic. Can't talk about dog."
        }
    ```

    Example 2: 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 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 
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
@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