Overview

Set of AI-powered tools for everyday use with OpenAi, Azure OpenAI or Anthropic LLMs.

1. Snippets — the actions that can be performed on selected text.
2. Assistants — your own specialized assistants to talk with.
3. Tools — your own specialized tools to use with Assistants.

KrAIna can be easily extended by personal, third-party sets of the above beings by creating folder (or create symlink) in KrAIna.

1. The folder must contain file named .kraina-land - this is tag file for KrAIna to scan the folder
2. To extend KrAIna with new snippets, assistants or tools create the respective folders names
3. Follow KrAIna structure of these new sets
4. Example folder structure

    ├── kraina
            ├──kraina-personal
                   ├── .kraina-land
                   ├── snippets
                   │     ├── create_jira 
                   │          ├── prompt.md
                   │          ├── config.yaml
                   ├── assistants
                   │        ├── database 
                   │            ├── prompt.md
                   │            ├── config.yaml
                   ├── tools
                   │     ├── database 
                   │          ├── __init__.py
                   │          ├── ...
                   │     ├── include.py
   

Chat - Chat GUI application built using tkinter for Assistants and Snippets.

Currently on available on Linux

Snippets

Snippets are actions that can be performed on selected text.

KrAIna can be easily equipped with new snippets. Check the snippets folder. The structure is as follows:

snipptes/
├── fix
│     ├── prompt.md - snippet system prompt, required
│     ├── config.yaml - snippet and LLM settings, optional
│     ├── py_module.py - overwrite default behavior of snippet, specialization - must be defined in model.yaml

config.yaml schema:

# Defaults are listed here which are used when config is not defined (not available in snippet config.yaml)
# Optional. Force api: azure or openai or anthropic
force_api: ""
model: gpt-3.5-turbo
temperature: 0.5
max_tokens: 512

You can add additional context to the Snippet (the same scheme as for Assistant) by adding:

contexts:
  string: Examples
  file:
    - ./example1.txt
    - ./example2.txt

to config.yaml file. The fields are parsed and added to system prompt:

... system prompt ...
Take into consideration the context below while generating answers.
# Context:
## 0
Examples
## 1
./example1.txt content with blocked template placeholders ({placeholder} -> {{placeholder}})
## 2
./example2.txt content with blocked template placeholders ({placeholder} -> {{placeholder}})
## 3
Current date: {date}

Note: This part is always added, regardless of whether the contexts key exists in config.yaml. This part is not added when the config.yaml file does not exist.

Take into consideration the context below while generating answers.
# Context:
## 0
Current date: {date}

However, AI-powered snippets are nothing without a good user interface to make it possible to use them in any tool. One way to boost your work performance is by performing snippets on the clipboard context with a Clipboard manager.

Assistants

Your personal AI assistant. It can be a causal assistant or prompt engineer or storyteller. Assistant can be run as one-shot, similar to snippets or can use its memory and remember the conversation.

The assistants have been designed similar to Snippets. Check the assistants folder.

config.yaml schema:

# Defaults are listed here which are used when config is not defined (not available in assistant config.yaml)
# Optional. Force api: azure or openai or anthropic
force_api: ""
model: gpt-3.5-turbo
temperature: 0.7
max_tokens: 512
tools:
  - 
description: 
contexts:
  # string: We are located in Haiti
  # string: 
  #   - We are located in Haiti
  #   - We have sunny whether
  # file: ./about_me.txt
  # file:
  #   - ./about_me.txt
  #   - ./my_projects.txt 

You can add additional context to the Assistant by adding:

contexts:
  string_template: We are located in {place} # Note that this string will be added as `We are located in {place}` (place placeholder must be provided)
  string: We are located in {place} # Note that this string will be added `as We are located in {{place}}` (with blocked template placeholder) 
  # string: 
  #   - We are located in Haiti
  #   - We have sunny whether
  file: ./about_me.txt
  file_template: ./about_me.txt
  # file:
  #   - ./about_me.txt
  #   - ./my_projects.txt 

to config.yaml file. The fields are parsed and added to system prompt:

... system prompt ...
Take into consideration the context below while generating answers.
# Context:
## 0
We are located in {place}
## 1
We are located in {{place}}
## 2
./about_me.txt content
## 3
Current date: {date}

Note: This part is always added, regardless of whether the contexts key exists in config.yaml. This part is not added when the config.yaml file does not exist.

Take into consideration the context below while generating answers.
# Context:
## 0
Current date: {date}

The assistants can use tools. To do this:

1. Assign tools (LangChain BaseTools) by listing them in assistant config.yaml key

   tools:
     - brave_web
     - file_search

2. Use models capable to do Functional Calling like gpt-4o, gpt-3.5-turbo, gpt-4-turbo

Tools

Set of tools available to others in krAIna.

To make such a tool, you need to follow these steps:

1. Find or develop a tool derived from BaseTool.
   1. Check https://python.langchain.com/v0.2/docs/integrations/tools/ for build-in in langchain tools
   2. Check https://python.langchain.com/v0.2/docs/integrations/toolkits/ for build-in in langchain tools
   3. Check https://python.langchain.com/v0.2/docs/how_to/custom_tools/ how to create your own tool
2. Create an initialization function that:
   1. Must accept one parameter, tool_settings (even if you don't have any settings).
   2. Must return BaseTool or List[BaseTool].
3. Add your tool to the SUPPORTED_TOOLS dictionary in tools/include.py. The name of your tool is the key of the SUPPORTED_TOOLS dictionary.

The initialization of the tool (calling the init function) occurs when an Assistant is called, not when it is initialized.

Chat GUI application

Chat GUI application build using tkinter.

features:

• Chat with history (html and text tabs available)
• Light/Dark theme
• Chats history which can be recalled. They are auto-named and describe
• Chat history management (Pin/Unpin, make inactive, edit name, description, delete permanent)
• Assistant selection
• Support for snippets — right-click in user query widget to apply transformation on a text
• Overwrite Assistant settings
• persistence storage on exit
• progress bar to visualize that LLM is working
• status bar with information about estimated token used for system prompt, history, tools, completions. OpenAI tiktoken is used. Thus, for Anthropic LLM, this can be less accurate
• Live token estimation for a user query. OpenAI tiktoken is used. Thus, for Anthropic LLM, this can be less accurate
• Inter-process communication. The chat app initiates an IPC host, enabling control, such as:
  • chat.sh SHOW_APP which run Chat application or show it. It can be assigned to a global shortcut in OS
  • run Chat application snippets chat.sh RUN_SNIPPET translate "Co tam słychać?"
• Markdown/HTML support
• Debug Window (left-bottom corner) with the application logs

Install

1. Clone the project.

2. Run the setup.sh script, which executes the commands below, or run the commands below by yourself.

   1. Create a virtual environment: python3 -m venv .venv
   2. Install the requirements from requirements.txt: pip install -r requirements.txt.
   3. Create a .env file (cp .env.template .env) and add:
      1. OPENAI_API_KEY=sk-... - OpenAI API key
      2. AZURE_OPENAI_ENDPOINT + AZURE_OPENAI_API_KEY + OPENAI_API_VERSION - AzureAI API key if you'd like to use it
      3. ANTHROPIC_API_KEY - Anthropic API key if you'd like to use it
      4. Tools providers API key
   4. Create a config.yaml (cp config.yaml.template config.yaml) and modify if needed.

---

Note: By default, the highest priority has Azure OpenAI LLM, next OpeAI and the last Anthropic. Thus, if all API keys exist, Azure OpenAI is selected. If OpenAI and Anthropic, OpenAi is selected.

Note: Versioning and auto-updating of the kraina.db schema are not supported at this time. if you had already created kraina.db, update schema:

create table messages_dg_tmp
(
    message_id      INTEGER  not null
        constraint pk_messages
            primary key,
    conversation_id INTEGER  not null
        constraint fk_messages_conversation_id_conversations
            references conversations
            on delete cascade,
    type            INTEGER  not null,
    message         VARCHAR  not null,
    create_at       DATETIME not null
);

insert into messages_dg_tmp(message_id, conversation_id, type, message, create_at)
select message_id, conversation_id, type, message, create_at
from messages;

drop table messages;

alter table messages_dg_tmp
    rename to messages;

create index ix_messages_conversation_id
    on messages (conversation_id);

create index messages_message_id_index
    on messages (message_id);

alter table conversations
    add priority integer default 0 not null;

---

CopyQ Custom Action Installation

1. Edit and save the copyQ/ai_select.ini file:
   • Adjust the path ~/krAIna/kraina.sh to your needs.
   • Change or remove shortcuts if needed (global shortcut ALT+SHIFT+1, CopyQ shortcut ALT+RETURN).
2. Open CopyQ and go to Command/Global shortcuts... .
3. Select Load Commands... and import the copyQ/ai_select.ini file.

Check also other CopyQ Custom Actions in copyQ.

---

Note:

1. Tested with CopyQ 7.1.0 (8.0.0 has some problem with main window focus)
2. To get popup notifications (usually on errors), disable Use native notifications in CopyQ Preferences...

---

Configuration

The configuration is handled by config.yaml file.

File Schema:

llm:
  # LLM settings
  map_model:
    # map model names from snippet/assistant yaml files into models per API type
    # Using alias like `A` or `B`, you can quickly change API from OpenAI to Anthropic
    azure:
      A: gpt-4o
      B: gpt-35-turbo
      gpt-4-turbo: gpt-4-turbo-128k
      gpt-3.5-turbo: gpt-35-turbo
    openai:
      A: gpt-4o
      B: gpt-3.5-turbo
    anthropic:
      A: claude-3-5-sonnet-20240620
      B: claude-3-haiku-20240307
chat:
   # chat settings
   # Always start New Chat with selected assistant
   default_assistant: samantha
   # maximum last chats to display in left sidebar
   visible_last_chats: 10
assistants:
   # assistants settings
   # assistant name:
   #    settings
   # aren't implemented yet
snippets:
   # snippets settings
   # snippet name:
   #    settings
   # aren't implemented yet
tools:
   # tools settings
   # tool name:
   #    settings
  brave_web:
    count: 3

Usage

CLI

1. Get all supported snippets: ./kraina.sh
2. Translate: ./kraina.sh translate "Cześć, co słychać u Ciebie?"
3. Git commit: ./kraina.sh commit "$(git diff --staged --no-prefix -U10)"

CopyQ Usage

To use the krAIna CopyQ Custom Action ai:select:

1. Select text.
2. Press ALT+SHIFT+1.
3. Select the snippet you'd like to use and press ENTER.
4. Once the action finishes, the selected text is replaced with the transformed one.

Alternatively:

1. Select and copy text to the clipboard.
2. Open CopyQ.
3. Right-click on the copied text and select the ai:select Custom Action (or press ALT+RETURN).
4. Once the action finishes, the selected text is replaced with the transformed one.

Chat

1. Start the application by running ./chat.sh.
2. Use its features.
3. You can also use ./chat.sh COMMAND to control the application with the following supported commands:

SHOW_APP - Trigger to display the application
HIDE_APP - Trigger to minimize the application
No argument - Run the GUI app. If the app is already running, it will be shown

Assign ./chat.sh SHOW_APP to system global shortkey to summon KrAIna Chat quickly, press Escape in application to minimalize it.

Code

Snippets

from dotenv import load_dotenv, find_dotenv
from snippets.base import Snippets

load_dotenv(find_dotenv())
snippets = Snippets()
action = snippets["fix"]
print(action.run("I'd like to speak something interest"))

Assistants

from dotenv import load_dotenv, find_dotenv
from assistants.base import Assistants

load_dotenv(find_dotenv())
assistants = Assistants()
# one shot, do not use a database
action = assistants["echo"]
ret = action.run("2+2", use_db=False)
print(ret)  # AssistantResp(conv_id=None, content='2 + 2 equals 4.', tokens={'api': {'model': 'gpt-3.5-turbo', 'max_tokens': 512, 'temp': 0.7}, 'prompt': 31, 'history': 0, 'input': 6, 'total_input': 37, 'output': 11, 'total': 85}, error=None)
# with history
first = action.run("My name is Paul")  # First call without conv_id creates new conversation
print(first)  # AssistantResp(conv_id=192, content='Nice to meet you, Paul! How can I assist you today?', tokens={'api': {'model': 'gpt-3.5-turbo', 'max_tokens': 512, 'temp': 0.7}, 'prompt': 31, 'history': 7, 'input': 7, 'total_input': 45, 'output': 17, 'total': 107}, error=None)
ret = action.run("What's my name?", conv_id=first.conv_id) # Second call with conv_id
print(ret)  # AssistantResp(conv_id=192, content='Your name is Paul. How can I assist you today, Paul?', tokens={'api': {'model': 'gpt-3.5-turbo', 'max_tokens': 512, 'temp': 0.7}, 'prompt': 31, 'history': 24, 'input': 8, 'total_input': 63, 'output': 17, 'total': 143}, error=None)