1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
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
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168

use std::sync::Arc;

use async_trait::async_trait;
use serde_json::json;

use crate::{
    agent::{agent::Agent, chat::prompt::FORMAT_INSTRUCTIONS, AgentError},
    chain::chain_trait::Chain,
    message_formatter,
    prompt::{
        HumanMessagePromptTemplate, MessageFormatterStruct, MessageOrTemplate, PromptArgs,
        PromptFromatter,
    },
    prompt_args,
    schemas::{
        agent::{AgentAction, AgentEvent},
        messages::Message,
    },
    template_jinja2,
    tools::Tool,
};

use super::{output_parser::ChatOutputParser, prompt::TEMPLATE_TOOL_RESPONSE};

pub struct ConversationalAgent {
    pub(crate) chain: Box<dyn Chain>,
    pub(crate) tools: Vec<Arc<dyn Tool>>,
    pub(crate) output_parser: ChatOutputParser,
}

impl ConversationalAgent {
    pub fn create_prompt(
        tools: &[Arc<dyn Tool>],
        suffix: &str,
        prefix: &str,
    ) -> Result<MessageFormatterStruct, AgentError> {
        let tool_string = tools
            .iter()
            .map(|tool| format!("> {}: {}", tool.name(), tool.description()))
            .collect::<Vec<_>>()
            .join("\n");
        let tool_names = tools
            .iter()
            .map(|tool| tool.name())
            .collect::<Vec<_>>()
            .join(", ");

        let sufix_prompt = template_jinja2!(suffix, "tools", "format_instructions");

        let input_variables_fstring = prompt_args! {
            "tools" => tool_string,
            "format_instructions" => FORMAT_INSTRUCTIONS,
            "tool_names"=>tool_names
        };

        let sufix_prompt = sufix_prompt.format(input_variables_fstring)?;
        let formatter = message_formatter![
            MessageOrTemplate::Message(Message::new_system_message(prefix)),
            MessageOrTemplate::MessagesPlaceholder("chat_history".to_string()),
            MessageOrTemplate::Template(
                HumanMessagePromptTemplate::new(template_jinja2!(
                    &sufix_prompt.to_string(),
                    "input"
                ))
                .into()
            ),
            MessageOrTemplate::MessagesPlaceholder("agent_scratchpad".to_string()),
        ];
        return Ok(formatter);
    }

    fn construct_scratchpad(
        &self,
        intermediate_steps: &[(AgentAction, String)],
    ) -> Result<Vec<Message>, AgentError> {
        let mut thoughts: Vec<Message> = Vec::new();
        for (action, observation) in intermediate_steps.into_iter() {
            thoughts.push(Message::new_ai_message(&action.log));
            let tool_response = template_jinja2!(TEMPLATE_TOOL_RESPONSE, "observation")
                .format(prompt_args!("observation"=>observation))?;
            thoughts.push(Message::new_human_message(&tool_response));
        }
        Ok(thoughts)
    }
}

#[async_trait]
impl Agent for ConversationalAgent {
    async fn plan(
        &self,
        intermediate_steps: &[(AgentAction, String)],
        inputs: PromptArgs,
    ) -> Result<AgentEvent, AgentError> {
        let scratchpad = self.construct_scratchpad(&intermediate_steps)?;
        let mut inputs = inputs.clone();
        inputs.insert("agent_scratchpad".to_string(), json!(scratchpad));
        let output = self.chain.call(inputs.clone()).await?.generation;
        let parsed_output = self.output_parser.parse(&output)?;
        Ok(parsed_output)
    }

    fn get_tools(&self) -> Vec<Arc<dyn Tool>> {
        self.tools.clone()
    }
}

#[cfg(test)]
mod tests {
    use std::{error::Error, sync::Arc};

    use async_trait::async_trait;
    use serde_json::Value;

    use crate::{
        agent::{chat::builder::ConversationalAgentBuilder, executor::AgentExecutor},
        chain::chain_trait::Chain,
        llm::openai::{OpenAI, OpenAIModel},
        memory::SimpleMemory,
        prompt_args,
        tools::Tool,
    };

    struct Calc {}

    #[async_trait]
    impl Tool for Calc {
        fn name(&self) -> String {
            "Calculator".to_string()
        }
        fn description(&self) -> String {
            "Usefull to make calculations".to_string()
        }
        async fn run(&self, _input: Value) -> Result<String, Box<dyn Error>> {
            Ok("25".to_string())
        }
    }

    #[tokio::test]
    #[ignore]
    async fn test_invoke_agent() {
        let llm = OpenAI::default().with_model(OpenAIModel::Gpt4.to_string());
        let memory = SimpleMemory::new();
        let tool_calc = Calc {};
        let agent = ConversationalAgentBuilder::new()
            .tools(&[Arc::new(tool_calc)])
            .build(llm)
            .unwrap();
        let input_variables = prompt_args! {
            "input" => "hola,Me llamo luis, y tengo 10 anos, y estudio Computer scinence",
        };
        let executor = AgentExecutor::from_agent(agent).with_memory(memory.into());
        match executor.invoke(input_variables).await {
            Ok(result) => {
                println!("Result: {:?}", result);
            }
            Err(e) => panic!("Error invoking LLMChain: {:?}", e),
        }
        let input_variables = prompt_args! {
            "input" => "cuanta es la edad de luis +10 y que estudia",
        };
        match executor.invoke(input_variables).await {
            Ok(result) => {
                println!("Result: {:?}", result);
            }
            Err(e) => panic!("Error invoking LLMChain: {:?}", e),
        }
    }
}