% DV2(1) DéjàVu-DéjàVu Chatbot/Scriptbot | Version 0.4.20 % Gary Dean % Oct 2023
DéjàVu-DéjàVu Chatbot/Scriptbot Terminal
dv2 [agent] [-options] [<script_file]
agent
Agent file (with '.dv2.agent' filename extension).
Optional, Positional; must be first argument,
before any other options.
script_file
script.dv2 file, via stdin. --
-v, --verbose : Verbose on (default)
-m, --model "model" : openai [model=gpt-3.5-turbo-16k-0613]
-t, --temperature N.N : [temperature=0.0072973525693]
-g, --message system|user|assistant "content..." : eg, -g user "How did I get here?"
-k, --token_limit N : context window [token limit=16000]\n. Usually calculated automatically (see --auto-max-tokens).
-M, --max_tokens N : response tokens [max_tokens=4000] Usually calculated automatically (see --auto-max-tokens).
-c, --command, --cmd { "/cmd" | "gpt_instruction" } : Add dv2 /slash command or gpt user prompt to command stack (Cmd_Cache[]). Can be used multiple times.
-L, --languages "lang1,lang2,..." : [languages='English Indonesian']
-x, --exit_on_completion, --exit : [exit_on_completion='0'] Exit upon completion of commands/script.
-v, --verbose : verbose=1 [verbose=1]
-q, --quiet : verbose=0
-N, --no-statusline : statusline=0 [statusline=1]
-i, --interactive : interactive=1 [interactive=1] yes-no prompts are enabled.
-y, --not-interactive : interactive=0 yes-no prompts are disabled; default y.
-a, --auto-max-tokens : auto_max_tokens=1 [auto_max_tokens=1] Enable dynamic calculation of max_tokens to maximum available. -A, --no-auto-max-tokens : auto_max_tokens=0 Disable dynamic calculation of max_tokens.
-a, --autosave on|off : If on, upon exit, append current messages to current dv script. Default is off.
-u, --upgrade : Upgrade DV2 from git repository. Git repository is set to: https://github.com/Open-Technology-Foundation/dejavu2
-V, --version : [version="dv2 0.4.20"], exit
-h, --help : This help, exit
_///// _// _//
_// _// _// _// _///_/
_// _// _// _// _/ _//
_// _// _// _// _//
_// _// _// _// _//
_// _// _//// _//
_///// _// _////////
D E J A V U - D E J A V U
DéjàVu-DéjàVu (DV2) is an AI Agent-creation and scripting program for the Bash terminal, using openAI GPT models.
DV2 has powerful scripting capabilities. It brings the power of OpenAi's GPT models directly to your terminal command-line.
__"it's dejavu all over again."__ - Y. Berra
Remembers context and history. You and your AI can have multiple 'personalities', 'experts', tasks, etc.
Especially useful for multi-chain "chain-of-thought" processes.
For the moment, DV2 has only been tested on Ubuntu Linux 22.04.
Other requirements are:
- Python 3.10+
- Bash 5+
- git
- pandoc
- openai
Before starting, you will need ensure you have an API key from openAI in order to run this program. See also ENVIRONMENT.
Installation One-Liner, if you're in a hurry:
git clone https://github.com/Open-Technology-Foundation/dejavu2 /tmp/dejavu2 && /tmp/dejavu2/dv2.installdv2.install will detect these variables during installation. If it doesn't, check your environment.
Once that's done, you're ready to install. dv2.install will:
- Execute
apt updateandapt upgrade(disable with --no-apt) - Install python 3, pip and git packages (disabled with --no-apt)
- Modify
~/.bashrcand/etc/bash.bashrcto include openai environment variables. - Store program files in
/usr/share/dejavu2 - Create symlink
dv2in/usr/local/bin.
Before running, make sure you have set up your openAI API key in your system's environment. If you set up your openAI account as an organization, you will also need to set your organization ID. Update your environment variables as follows:
export OPENAI_API_KEY='sk-_your_key_'
export OPENAI_ORGANIZATION_ID='org-_your_org_id_'You may wish to place these declarations into your .bash.rc and/or /etc/bash.bashrc files.
# Run interactively with default config Agents
dv2
# Add agent 'mydharma', interactive
dv2 mydharma
# Add agent 'techlead', ask question, and exit.
dv2 techlead -xc 'in python, display syntax, options, usage and examples for .replace()'-
Config: Default configuration settings for DV2 are loaded serially from Agent file/s.
-
Script: If there is input from stdin from a *.DV2 script file, this is pre-loaded into the DV2 command stack.
-
Command-line: parameters are processed.
-
DV2 Stack Execution: If the DV2 command stack is not empty, execution of command stack commences.
-
DV2 Command Prompt: Manually enter /cmds or instructions to GPT.
Upon execution, DV2 attempts to load DV2.conf files from the following locations, in this order:
- ${script_dir}/DV2.conf
- /etc/default/DV2.conf
- $HOME/.DV2.conf
- ./.DV2.conf
If an Agent file has been specified on the command line this is loaded last.
DV2.conf files found on this execution are:
${CONF_FILES[*]}
Each Agent inherits all the setting from the previous Agent, thus each subsequent Agent has the ability to set some or all of the available parameters.
DV2 script files have the extension .DV2, and comprises /commands and gpt queries.
DV2 script commands are differentiated from GPT instructions by starting with a '/' character (or alternatively, a '!' character).
For instance, /status will display all current DV2 settings.
range can be in the forms "1,2,3", "4-6", "7-", "-8", "all" and can be combined in any order.
/ Short command help.
/help Open DV2 help file. Can also use "//"
/exit Exit DV2. Pressing ^C will also exit.
/status Show status of current environment.
/vars Show current variables.
/list [long|short] [range]
List current messages.
"short" for condenced list, "long" for full list.
Default is "long".
If "range" omitted, lists entire messages.
/delete range Delete message items in "range".
/clear Clear all messages. Same as /delete 1-
/tldr [range] Summarize all messages responses in "range".
Default is the previous response.
/summarize [messages|prompt|all]
Summarise every message or prompt items.
Default is "messages".
/autosave [on|off]
Save all message to current dv Script on exit.
Default is Off.
/save [file] Save current message.
/import [file] Import "file" into the input prompt.
If "file" is not specified, opens EDITOR to enable
multi-line commands.
/user_name [name] Set/Display user name.
/ai_name [name] Set/Display AI name.
/engine [list|engine|update]
Set/Display GPT engine.
/temperature [f] Set/Display temperature (0.0-1.0).
/top_p [f] Set/Display top_p (0.0-1.0).
/tokens [n] Set/Display tokens to use.
/prompt [prompt] Display current messages set-up information.
If "prompt" is specified, set the new conversion prompt.
/files Display messages scripts in current and user home
directories, with option to edit.
/edit [file] Edit "file". If "file" not specified, edit the current
script file.
/run [file] Run the "file" script. If not specified, display a list
of available scripts.
/instruction string
Insert an instruction. Usually only used in scripts.
/messages string
Add to messages array. Usually only used in scripts.
/echo [on|off] Turn command echo on|off. Usually only used in scripts.
/exec [cmd...] Execute a shell command.
Prompt Engineering is a critial part of getting the most out of DV2 and GPT models. Here are some guidelines:
The most important factor to consider when designing a prompt is context. Making sure the context is relevant is crucial for getting coherent and accurate responses from GPT.
Without sufficient context, GPT may generate responses that are off-topic, irrelevant, or inconsistent with the goal of the prompt. To ensure the prompt has an adequate amount of context, include all relevant background information.
After providing context, the next step to designing an effective prompt is to define a clear instruction for GPT.
This requires that you have a clear understanding of the task to be completed, and the task definition should be specific, concise, and avoid ambiguity or vagueness.
When designing a prompt make sure the prompt is specific. The more details and precision included in the prompt, the more likely it is that the GPT will generate a targeted and accurate response.
This includes important details such as what the goal is, the starting and ending points, characters involved, or any relevant background information. If the prompt is too vague, it will result in off-topic, irrelevant, or inconsistent responses.
Iteration is an effective way of designing an effective prompt. Prompt design is often an iterative process that involves multiple attempts and cycles of design, testing, and evaluation.
Each iteration offers an opportunity to refine or improve the prompt. For example, if GPT generates an off-topic response, you could add more specific instructions or additional context to the prompt.
The iterative approach it allows for continuous improvement and optimization of the generated content.
Python 3, pip, git, openai API key/s, apt install access
Report bugs and deficiencies on the DV2 github page
Copyright © 2022-2023 Indonesian Open Technology Foundation. License GPLv3+: GNU GPL version 3 or later GNU Licences. This is free software: you are free to change and redistribute it. There is NO WARRANTY, to the extent permitted by law.