DolphSa LM v1.7

Overview

DolphSa is a complex of uncensored, heavily customizable lightweight language models designed for offline local execution on consumer-grade hardware.

The models that come for free are DolphSa and DolphSa-ultralight. They can only be used through the GUI and the CLI. Unlike the pro models, they are not designed to be embedded within applications, even though you technically can by making your program run shell commands to interact with the model/s.

Graphical User Interface (GUI)

Open it like a standard application.

On the bottom, you have the text box where you can type your prompt. On the left side, you have your chats. On the bottom left side, you have the setting to switch between light/dark mode. Right below it, you have the settings menu. Clicking it opens the following configuration options:

On the top right corner, you can see a trash icon. This will delete and overwrite the current conversation. Once performed, it cannot be recovered, so be careful.

Conversations are stored in %temp%. If you ask the model to write code, it may write it to your current working directory. It will ask for permission first; in the GUI, you will receive a popup asking if you want to give the model permission to write files in that specific directory. Note that this will not always happen when it writes code, it only occurs in cases of full-code generation. Pseudocode or snippets will be stored directly along with the conversation history.

Command Line Interface (CLI)

Type dolphsa to start it in your favorite shell (CMD, PowerShell, or any standard terminal).

Type dolphsa-cfg to open the config file and adjust settings with your preferred IDE or text editor.

The CLI works exactly as in the GUI version. When writing code, it may prompt for permission to write files to a specific directory with a Y/N prompt.

Pro Tier & Embedded Integration

The Pro version is essentially the same, but includes much higher-performing models along with built-in, reliable ways to embed the engine directly into your applications.

Available Pro models:

The system can be used with any programming language easily. You can start a background process and communicate with it using a pipe. It includes safety measures to prevent unverified programs from communicating instructions to it; you can turn this off, but only do so for local testing reasons as it poses a security risk otherwise. There is also an optional API key setting, only used for adding automatic-update support to our model when embedded within your app.

SDK Code Examples

Python Example [py-code]

import dolphsa

# Initialize model (model_name, api_key)
ai = dolphsa.Mdl("dolphsa-compact", "wEmdHUxV4zzv9eWr8J332")

# ask: takes prompt, returns text response
response = ai.ask("Brainstorm 5 creative ideas")
print(response)

# torf: takes prompt, strictly returns True or False
is_valid = ai.torf("Is oxygen required for human respiration?")
print("True or False:", is_valid)

# writefile: takes file path and prompt, writes generated output directly
ai.writefile("./output.txt", "Generate a detailed project log header.")

# context: adds persistent context for the duration of the current session
ai.context("Project context: Target audience is software engineers.")

# savecontext: saves active context state to %temp% for persistence across re-opens
ai.savecontext()

# censor: enables model censorship (not 100% immune to jailbreaks)
ai.censor()

# clearcontext: clears session context and resets censorship state
ai.clearcontext()

Go Example [go-code]

package main

import (
    "fmt"
    "example/dolphsa"
)

func main() {
    // Initialize model
    ai := dolphsa.NewMdl("dolphsa-compact", "wEmdHUxV4zzv9eWr8J332")

    // ask: takes prompt, returns text response
    response := ai.Ask("Brainstorm 5 creative ideas")
    fmt.Println(response)

    // torf: takes prompt, strictly returns true/false boolean
    isTrue := ai.Torf("Is water liquid at room temperature?")
    fmt.Println("Result:", isTrue)

    // writefile: takes file path and prompt
    ai.Writefile("./build.log", "Write a summary of current build tasks.")

    // context: adds persistent context for active session
    ai.Context("System context: Keep answers technical and concise.")

    // savecontext: saves active context to %temp%
    ai.Savecontext()

    // censor: enables output censorship
    ai.Censor()

    // clearcontext: clears context and resets censor settings
    ai.Clearcontext()
}

C# Example [c#-code]

using System;
using dolphsaEngine;

class Program
{
    static void Main()
    {
        // Initialize model
        var ai = new dolphsa.Mdl("dolphsa-compact", "wEmdHUxV4zzv9eWr8J332");

        // ask: takes prompt, returns string response
        string response = ai.Ask("Brainstorm 5 creative ideas");
        Console.WriteLine(response);

        // torf: takes prompt, strictly returns boolean true/false
        bool result = ai.Torf("Is light faster than sound?");
        Console.WriteLine($"Answer: {result}");

        // writefile: takes file path and prompt
        ai.Writefile("./notes.txt", "Draft an introduction for a user manual.");

        // context: adds persistent context to the current session
        ai.Context("Session context: Respond in standard Markdown format.");

        // savecontext: saves context state to %temp%
        ai.Savecontext();

        // censor: enables censorship filters
        ai.Censor();

        // clearcontext: clears session context and removes censorship status
        ai.Clearcontext();
    }
}