Skip to content

architechture.md

Latest commit

 

History

History
64 lines (51 loc) · 3.25 KB

architechture.md

File metadata and controls

64 lines (51 loc) · 3.25 KB

Bot

Bot will accept "name", "permission" and "system_prompt", "llm", "environment", "additional_tools", "folder_to_mount
    "Bot_type" => "READING_Bot" | "WRITING_Bot" | "BROWSING_Bot" | "CUSTOM_Bot"
    "permission" => "READ_ONLY" | "READ_WRITE" (optional field)
    "model_name" => { // Based on provider the Bot will instantiate llm instance
        provider: "openai" 
        model: "gpt-4" | "gpt-3.5-turbo" | "claude",
    }
    <!-- LLM should not give any interactive commands like vim -->
    "system_prompt" => string (Optional argument)
        when not passed will create default system prompt based on Bot_type and permission
    "environment" => environment instance (Optional argument)
        when not passed will create docker environment with default arguments as per PERMISSION based on Bot_type
        and tools will be installed one by one after environment creation
    "additional_tools" => array of tool definitions (Optional argument)
        when passed will create docker environment with default arguments as per PERMISSION and with default "tool" installations with additional tools installed

Bot will expose "run" method which will accept a "task", "timeout" and "max_iterations"
    will call llm task as user_prompt and get response
    get the response and call environment execute method with the command
    get the output and call llm with previous_communications and get the response
    repeat the above steps until llm response is final answer or max_iterations reached or timeout reached
    ensure to log every communication between llm and environment

    <!-- In the end llm should provide status "DONE" status in one object property with results in another property -->

    will return { status: bool, result: string, error: string | None, log: log of communication between llm and environment }

environment

environment will accept "folder_to_mount", "permission"
    "folder_to_mount" => string (path to the folder to mount inside docker)
    "permission" => "READ_ONLY" | "READ_WRITE"

environment will expose "execute" method which will accept command, timeout and return {
    return_code: int,
    output: string,
    error: string | None
}
    will execute the command inside the docker with a default timeout and return the output
environment will expose "kill" method which will kill the environment
    will stop the docker container and remove it

llm class

llm will accept "system_prompt" and "deployment_name"
system_prompt => string, 
deployment_name => string (like my-custom-gpt5-Bot)

llm will expose ask method which will accept "message" and will return llm response in particular json format
    the method will append each message to messages array and each response also will be appended to the messages array, 
llm will expose clear history method which will clear the messages array except the system prompt

tool definition singleton class

tool definition will accept "installation_command", "verification_command", "usage_instructions"

Whenever I am writing code

import logging
logger = logging.getLogger(__name__)
logger.info("This is an info message")