ambi 0.3.5

A flexible, multi-backend, customizable AI agent framework, entirely based on Rust.
Documentation 
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

use ambi::llm::providers::openai_api::config::OpenAIEngineConfig;
use ambi::types::ToolCallParser;
use ambi::{Agent, LLMEngineConfig};
use anyhow::Result;
use serde_json::Value;

// Step 1: Define the struct for your custom XML syntax parser.
pub struct XmlToolParser;

// Step 2: Implement the `ToolCallParser` trait.
impl ToolCallParser for XmlToolParser {
    fn get_tags(&self) -> (String, String) {
        // Return the opening and closing tags for the XML format.
        unimplemented!()
    }

    // Define how to instruct the LLM to format its tool calls.
    // This dynamically generates the tool instructions appended to the system prompt.
    fn format_instruction(&self, tools_json: &str) -> String {
        format!(
            "You can use tools. Call format:\n<tool_call name=\"tool_name\">{{\"args\":{{...}}}}</tool_call>\nAvailable tools:\n{}",
            tools_json
        )
    }

    // Define the logic to extract tool names and JSON arguments from the raw LLM output text.
    fn parse(&self, text: &str) -> Vec<(String, Value)> {
        let mut calls = Vec::new();
        let start_tag = "<tool_call name=\"";
        let end_tag = "</tool_call>";

        let mut current_text = text;

        // Iterate through the text to find all occurrences of the XML tags.
        while let Some(start_idx) = current_text.find(start_tag) {
            let name_start = start_idx + start_tag.len();

            // Extract the tool's name from the XML attribute.
            if let Some(quote_idx) = current_text[name_start..].find("\">") {
                let tool_name = &current_text[name_start..name_start + quote_idx];
                let content_start = name_start + quote_idx + 2;

                // Extract the JSON payload situated between the opening and closing tags.
                if let Some(end_idx) = current_text[content_start..].find(end_tag) {
                    let json_str = &current_text[content_start..content_start + end_idx];

                    // Attempt to parse the extracted string into a JSON Value.
                    if let Ok(args) = serde_json::from_str::<Value>(json_str) {
                        calls.push((tool_name.to_string(), args));
                    } else {
                        println!("Error: Invalid JSON arguments for tool '{}'", tool_name);
                    }
                    current_text = &current_text[content_start + end_idx + end_tag.len()..];
                } else {
                    break;
                }
            } else {
                break;
            }
        }
        calls
    }
}

#[tokio::main]
async fn main() -> Result<()> {
    // Step 3: Configure the backend engine.
    let engine_config = LLMEngineConfig::OpenAI(OpenAIEngineConfig {
        api_key: "mock-key".to_string(),
        base_url: "https://api.openai.com/v1".to_string(),
        model_name: "gpt-4o-mini".to_string(),
        temp: 0.7,
        top_p: 0.9,
    });

    // Step 4: Instantiate the Agent and inject the custom tool parser using `.with_tool_parser()`.
    // The framework will now instruct the LLM to output XML and parse it accordingly.
    let _agent = Agent::make(engine_config)
        .await?
        .with_tool_parser(XmlToolParser);

    Ok(())
}