daggerai/dist/src/core/parser.js

A simple and powerful Typescript based agent framework to help businesses thrive in the AI Agent revolution.

"use strict";
Object.defineProperty(exports, "__esModule", { value: true });
exports.ParserError = exports.AgentStep = exports.AgentAction = exports.AgentFinish = exports.SquadResponseParser = void 0;
const FINAL_ANSWER = 'Final Answer:';
const ERROR_MISSING_ACTION_AFTER_THOUGHT = "I made a mistake providing an invalid format: I missed the 'Action:' after 'Thought:'. I will fix it next time, and I will not use a tool I have already used.\n";
const ERROR_MISSING_ACTION_INPUT_AFTER_ACTION = "I made a mistake providing an invalid format: I missed the 'Action Input:' after 'Action:'. I will fix it next time, and don't use a tool I have already used.\n";
const ERROR_FINAL_ANSWER_AND_PARSABLE_ACTION = 'I made a mistake, I tried both performing an Action and giving a Final Answer, I must choose one or the other. I will fix it next time.\n';
class SquadResponseParser {
    emitter = null;
    parse(response) {
        const includesFinalAnswer = response.includes(FINAL_ANSWER);
        const hasActionAndInput = /Action\s*\d*\s*:[\s]*(.*?)[\s]*Action\s*\d*\s*Input\s*\d*\s*:[\s]*(.*)/g.test(response);
        if (hasActionAndInput && includesFinalAnswer) {
            throw new ParserError('Error parsing LLM ouput.', ERROR_FINAL_ANSWER_AND_PARSABLE_ACTION);
        }
        if (includesFinalAnswer) {
            const finalAnswer = response.split(FINAL_ANSWER)[1];
            if (finalAnswer) {
                return new AgentFinish(finalAnswer.trim());
            }
        }
        if (hasActionAndInput) {
            const actionRegex = /Action:\s*(.*?)\s*\nAction Input/;
            const actionInputRegex = /Action Input:\s*([\s\S]+)/;
            let actionMatch = response.match(actionRegex);
            const action = actionMatch?.[1]?.trim();
            let actionInputMatch = response.match(actionInputRegex);
            const actionInput = actionInputMatch?.[1]?.trim() || '';
            if (actionInput) {
                const jsonActionInput = JSON.parse(actionInput.trim());
                return new AgentStep(new AgentAction(action, jsonActionInput, response));
            }
        }
        const hasAction = /Action:\s*(.*?)\s*/.test(response);
        const hasActionInput = /[\s]*Action\s*\d*\s*Input\s*\d*\s*:[\s]*(.*)/.test(response);
        if (!hasAction) {
            throw new ParserError('Error parsing LLM ouput.', ERROR_MISSING_ACTION_AFTER_THOUGHT);
        }
        if (hasActionInput && !hasAction) {
            throw new ParserError('Error parsing LLM ouput.', ERROR_MISSING_ACTION_INPUT_AFTER_ACTION);
        }
        const finalAnswer = response.split(FINAL_ANSWER)[1];
        return new AgentFinish(finalAnswer?.trim() || 'It was not possible to execute the task.');
    }
}
exports.SquadResponseParser = SquadResponseParser;
class AgentFinish {
    response;
    constructor(response) {
        this.response = response;
    }
}
exports.AgentFinish = AgentFinish;
class AgentAction {
    tool;
    toolInput;
    log;
    constructor(tool, toolInput, log) {
        this.tool = tool;
        this.toolInput = toolInput;
        this.log = log;
    }
}
exports.AgentAction = AgentAction;
class AgentStep {
    action;
    observation;
    constructor(action, observation) {
        this.action = action;
        this.observation = observation;
    }
}
exports.AgentStep = AgentStep;
class ParserError extends Error {
    observation;
    constructor(message, observation) {
        super(message);
        this.observation = observation;
    }
}
exports.ParserError = ParserError;
//# sourceMappingURL=parser.js.map