Skip to content

Latest commit

 

History

History
436 lines (322 loc) · 12.3 KB

README.md

File metadata and controls

436 lines (322 loc) · 12.3 KB
Raw

AIChat

CI Crates

All-in-one CLI tool for 10+ AI platforms, including OpenAI, Gemini, Claude, Mistral, LocalAI, Ollama, VertexAI, Ernie, Qianwen...

Command Mode:

command mode

Chat REPL mode:

chat-repl mode

Install

Use a package management tool

For Rust programmer

cargo install aichat

For macOS Homebrew or a Linuxbrew user

brew install aichat

For Windows Scoop user

scoop install aichat

For Android Termux user

pkg install aichat

Binaries for macOS, Linux, and Windows

Download it from GitHub Releases, unzip, and add aichat to your $PATH.

Features

  • Support most of the LLM platforms
    • OpenAI: GPT3.5/GPT4 (paid, vision)
    • Gemini (free, vision)
    • Claude: Claude2/Claude3 (paid)
    • Mistral (paid)
    • LocalAI (free, local, vision)
    • Ollama (free, local)
    • Azure-OpenAI (paid)
    • VertexAI: Gemini-1/Gemini-1.5 (paid, vision)
    • Ernie (paid)
    • Qianwen (paid, vision)
  • Support Command Mode and Chat-REPL Mode
  • Support roles
  • Support sessions (context-aware conversation)
  • Support multimodal models (vision)
  • Execute commands using natural language
  • Shell integration
  • Syntax highlighting for markdown and 200+ languages in code blocks
  • Save messages/sessions
  • Stream/Non-stream output
  • Custom theme
  • With proxy

Config

On first launch, aichat will guide you through the configuration.

> No config file, create a new one? Yes
> AI Platform: openai
> API Key: <your_api_key_here>

Feel free to adjust the configuration according to your needs.

model: openai:gpt-3.5-turbo      # LLM model
temperature: 1.0                 # GPT temperature, between 0 and 2
save: true                       # Whether to save the message
highlight: true                  # Set false to turn highlight
light_theme: false               # Whether to use a light theme
wrap: no                         # Specify the text-wrapping mode (no, auto, <max-width>)
wrap_code: false                 # Whether wrap code block
auto_copy: false                 # Automatically copy the last output to the clipboard
keybindings: emacs               # REPL keybindings. values: emacs, vi
prelude: ''                      # Set a default role or session (role:<name>, session:<name>)
compress_threshold: 1000         # Compress session if tokens exceed this value (valid when >=1000)

clients:
  - type: openai
    api_key: sk-xxx

  - type: localai
    api_base: http://localhost:8080/v1
    models:
      - name: gpt4all-j
        max_input_tokens: 8192

Take a look at the config.example.yaml for the complete configuration details.

There are some configurations that can be set through environment variables. For more information, please refer to the Environment Variables page.

Command

Usage: aichat [OPTIONS] [TEXT]...

Arguments:
  [TEXT]...  Input text

Options:
  -m, --model <MODEL>        Choose a LLM model
  -r, --role <ROLE>          Choose a role
  -s, --session [<SESSION>]  Create or reuse a session
  -e, --execute              Execute commands using natural language
  -c, --code                 Generate only code
  -f, --file <FILE>...       Attach files to the message to be sent
  -H, --no-highlight         Disable syntax highlighting
  -S, --no-stream            No stream output
  -w, --wrap <WRAP>          Specify the text-wrapping mode (no, auto, <max-width>)
      --light-theme          Use light theme
      --dry-run              Run in dry run mode
      --info                 Print related information
      --list-models          List all available models
      --list-roles           List all available roles
      --list-sessions        List all available sessions
  -h, --help                 Print help
  -V, --version              Print version

Here are some practical examples:

aichat                                          # Start in REPL mode

aichat -e install nvim                          # Execute
aichat -c fibonacci in js                       # Code

aichat -s                                       # REPL + New session
aichat -s session1                              # REPL + New/Reuse 'session1'

aichat --info                                   # System info
aichat -r role1 --info                          # Role info
aichat -s session1 --info                       # Session info

cat data.toml | aichat -c to json > data.json   # Pipe IO

aichat --file a.png b.png -- diff images        # Attach files

Execute commands using natural language

Simply input what you want to do in natural language, and aichat will prompt and run the command that achieves your intent.

aichat -s <text>...

Aichat is aware of OS and $SHELL you are using, it will provide shell command for specific system you have. For instance, if you ask aichat to update your system, it will return a command based on your OS. Here's an example using macOS:

aichat -e update my system
# sudo softwareupdate -i -a
# ? [e]xecute, [d]escribe, [a]bort:  (e)

The same prompt, when used on Ubuntu, will generate a different suggestion:

 aichat -e update my system
# sudo apt update && sudo apt upgrade -y
# ? [e]xecute, [d]escribe, [a]bort:  (e)

We can still use pipes to pass input to aichat and generate shell commands:

aichat -e POST localhost with < data.json
# curl -X POST -H "Content-Type: application/json" -d '{"a": 1, "b": 2}' localhost
# ? [e]xecute, [d]escribe, [a]bort:  (e)

We can also pipe the output of aichat which will disable interactive mode.

aichat -e find all json files in current folder | pbcopy

Shell integration

This is a very handy feature, which allows you to use aichat shell completions directly in your terminal, without the need to type aichat with prompt and arguments. This feature puts aichat completions directly into terminal buffer (input line), allowing for immediate editing of suggested commands.

aichat-integration

To install shell integration, go to ./scripts/shell-integration to download the script and source the script in rc file. After that restart your shell. You can invoke the completion with alt+e hotkey.

Generate Code

By using the --code or -c parameter, you can specifically request pure code output, for instance:

aichat --code a echo server in node.js

const net = require('net');

const server = net.createServer(socket => {
  socket.on('data', data => {
    socket.write(data);
  });

  socket.on('end', () => {
    console.log('Client disconnected');
  });
});

server.listen(3000, () => {
  console.log('Server running on port 3000');
});

Since it is valid js code, we can redirect the output to a file:

aichat --code a echo server in node.js > echo-server.js 
node echo-server.js

Chat REPL

aichat has a powerful Chat REPL.

The REPL supports:

  • Tab autocomplete
  • Custom REPL Prompt
  • Emacs/Vi keybinding
  • Edit/paste multiline text
  • Open an editor to modify the current prompt
  • Undo support

.help - print help message

> .help
.help                    Print this help message
.info                    Print system info
.model                   Switch LLM model
.role                    Use a role
.info role               Show role info
.exit role               Leave current role
.session                 Start a context-aware chat session
.info session            Show session info
.clear messages          Clear messages in the session
.exit session            End the current session
.file                    Attach files to the message and then submit it
.set                     Modify the configuration parameters
.copy                    Copy the last reply to the clipboard
.exit                    Exit the REPL

Type ::: to begin multi-line editing, type ::: to end it.
Press Ctrl+O to open an editor to modify the current prompt.
Press Ctrl+C to abort readline, Ctrl+D to exit the REPL

.info - view information

> .info
model               openai:gpt-3.5-turbo
temperature         -
dry_run             false
save                true
highlight           true
light_theme         false
wrap                no
wrap_code           false
auto_copy           false
keybindings         emacs
prelude             -
compress_threshold  1000
config_file         /home/alice/.config/aichat/config.yaml
roles_file          /home/alice/.config/aichat/roles.yaml
messages_file       /home/alice/.config/aichat/messages.md
sessions_dir        /home/alice/.config/aichat/sessions

.model - choose a model

> .model openai:gpt-4
> .model ollama:llama2

You can easily enter model name using autocomplete.

.role - let the AI play a role

Select a role:

> .role emoji

Send message with the role:

emoji> hello
👋

Leave current role:

emoji> .exit role

> hello
Hello there! How can I assist you today?

Show role info:

emoji> .info role
name: emoji
prompt: I want you to translate the sentences I write into emojis. I will write the sentence, and you will express it with emojis. I just want you to express it with emojis. I don't want you to reply with anything but emoji. When I need to tell you something in English, I will do it by wrapping it in curly brackets like {like this}.
temperature: null

Temporarily use a role to send a message.

> ::: .role emoji
hello world
:::
👋🌍

>

.session - context-aware conversation

By default, aichat behaves in a one-off request/response manner.

You should run aichat with -s/--session or use the .session command to start a session.

> .session

temp) 1 to 5, odd only                                                                    0
1, 3, 5

temp) to 7                                                                        19(0.46%)
1, 3, 5, 7

temp) .exit session                                                               42(1.03%)
? Save session? (y/N)

The prompt on the right side is about the current usage of tokens and the proportion of tokens used, compared to the maximum number of tokens allowed by the model.

.file - attach files to the message

Usage: .file <file>... [-- text...]

.file message.txt
.file config.yaml -- convert to toml
.file a.jpg b.jpg -- What’s in these images?
.file https://ibb.co/a.png https://ibb.co/b.png -- what is the difference?

Only the current model that supports vision can process images submitted through .file command.

.set - modify the configuration temporarily

.set temperature 1.2
.set compress_threshold 1000
.set dry_run true
.set highlight false
.set save false
.set auto_copy true

Roles

We can define a batch of roles in roles.yaml.

Retrieve the location of roles.yaml through the REPL .info command or CLI --info option.

For example, we can define a role:

- name: shell
  prompt: >
    I want you to act as a Linux shell expert.
    I want you to answer only with bash code.
    Do not provide explanations.

Let ChatGPT answer questions in the role of a Linux shell expert.

> .role shell

shell>  extract encrypted zipfile app.zip to /tmp/app
mkdir /tmp/app
unzip -P PASSWORD app.zip -d /tmp/app

AIChat with roles will be a universal tool.

$ aichat --role shell extract encrypted zipfile app.zip to /tmp/app
unzip -P password app.zip -d /tmp/app

$ cat README.md | aichat --role spellcheck

For more details about roles, please visit Role Guide.

License

Copyright (c) 2023-2024 aichat-developers.

Aichat is made available under the terms of either the MIT License or the Apache License 2.0, at your option.

See the LICENSE-APACHE and LICENSE-MIT files for license details.