Skip to content

Latest commit

 

History

History
132 lines (103 loc) · 6.07 KB

EXAMPLES.md

File metadata and controls

132 lines (103 loc) · 6.07 KB

Examples

All of the commands support xargs-like -i/-I/-replace flags.

Example: clai h | clai -i q Summarize this for me: {}, this would summarize the output of clai h.

Regardless of you wish to generate a photo, continue a chat or reply to your previous query, the prompt system remains the same.

clai help `# For more info about the available commands (and shorthands)`

Queries

clai query My favorite color is blue, tell me some facts about it
clai -re `# Use the -re flag to use the previous query as context for some next query` \
    q Write a poem about my favorite colour

Personally I have alias ask=clai q and then alias rask=clai -re q. This way I can ask -> rask -> rask for a temporary conversation.

Every 'temporary conversation' is also saved as a chat, so it's possible to continue it later, see below on how to list chats.

Tooling

Many vendors support function calling/tooling. This basically means that the AI model will ask your local machine to run some command, then it will analyze the output of said command.

See all the currently available tools here, please create an issue if you'd like to see some tool added.

clai -t q  `# Specify you wish to enable tools with -t/-tools` \
   Analyze the project found at ~/Projects/clai and give me a brief summary of what it does

Chatting

clai -chat-model claude-3-opus-20240229 `  # Using some other model` \
    chat new Lets have a conversation about Hegel

The -cm/-chat-model flag works for any text-like command. Meaning: you can start a conversation with one chat model, then continue it with another.

clai chat list
clai c continue Lets_have_a_conversation_about
clai c continue 1 kant is better `# Continue some previous chat with message `

Globs

clai -raw `# Don't format output as markdown` \
    glob '*.go' Generate a README for this project > README.md

The -raw flag will ensure that the output stays what the model outputs, without glow or animations.

Note that the glob mode also can be used by using the -g flag, as in, clai -g '<glob>' query/chat/photo/cmd. Glob-as-arg will be deprecated at some point.

Cmd

clai cmd to show all files in home

Will work like many of the popular command suggestion LLM tools out there. Flags works with this mode as well, such as clai -re -g 'some_file.go' cmd to cleanup this messy code, but it's not guaranteed the LLM will output an executable output.

Profiles

  1. clai setup -> 2 -> n
  2. Write some profile, example 'gopher'
  3. clai -p gopher -g './internal/moneymaker/handler_test.go' q Fix the tests in this file

Profiles allows you to preconfigure certain fields which will be passed to the llms, most noteably the prompt and which tools to use. This, in turn, enables you to quickly swap between different 'LLM-modes'.

For instance, you may have one profile which is prompted for golang programming tasks "gopher", it has tools write_file, rip grep and go enabled, and then another profile which is for terraform named "terry". With these, you don't have to 'pre-prompt' with clai q _in terraform_ ... or clai q _in golang_ ... but instead can use clai -p terry q .../clai -p gopher q ... and also restrict which tools are allowed, as opposed to enabling all tools (with -t).

These profiles are saved as json at os.GetConfigDir()/.clai/profiles. This means that you can sync them across all of your machines and tweak your prompts wherever you code.

Yet again, I've personally utilized aliases here. ask -> Generic profile-less prompt gask -> clai -p gopher q, grask -> clai -re -p gopher q and then task -> clai -p terry q, etc. These aliases are later on synched with the rest of my dotfiles + clai profiles, so they're shared on all my development machines.

Photos

printf "flowers" | clai -i `    # stdin replacement works for photos also` \
    --photo-prefix=flowercat `  # Sets the prefix for local photo` \
    --photo-dir=/tmp `          # Sets the output dir` \
    photo A cat made out of {}

Since -N alternatives are disabled for many newer OpenAI models, you can use repeater to generate several responses from the same prompt:

NO_COLOR=true repeater -n 10 -w 3 -increment -file out.txt -output BOTH \
    clai -pp flower_INC p A cat made of flowers

Configuration

clai will create configuration files at os.GetConfigDir()/.clai/. First time you run clai, two default command-related ones, textConfig.json and photoConfig.json, will be created and then one for each specific model. The configuration presedence is as follows (from lowest to highest):

  1. Default hard-coded configurations such as this, these gets written to file first time you run clai
  2. Configurations from local textConfig.json or photoConfig.json file
  3. Profiles
  4. Flags

The textConfig.json/photoConfig.json files configures what you want done, not how the models should perform it. This way it scales for any vendor + model.

Models

There's two ways to configure the models:

  1. Set flag -chat-model or -photo-model
  2. Set the model field in the textConfig.json or photoConfig.json file. This will make it default, if not overwritten by flags.

Then, for each model, a new configuration file will be created. Since each vendor's model supports quite different configurations, the model configurations aren't exposed as flags. Instead, modify the model by adjusting its configuration file, found in os.GetConfigDir()/.clai/<vendor>_<model-type>_<model-name>.json. This config json will in effect be unmarshaled into a request send to the model's vendor.

Conversations

Within os.GetConfigDir()/.clai/conversations you'll find all the conversations. You can also modify the chats here as a way to prompt, or create entirely new ones as you see fit.