Integrate AI into bash in a way you already use, with ctrl+r aka reverse-ai-search
Reverse AI Search provides local AI bash completion in a format you're already familiar with.
Only works with bash currently. Zsh coming soon.
# find all yaml files
(reverse-ai-search): '#find all yaml files' find . -name '*.yaml'
$ find . -name '*.yaml'
(reverse-ai-search): 'suod vmi /ect/hots' sudo vim /etc/hosts
$ sudo vim /etc/hosts
(reverse-ai-search): 'curl localhost | jq #last item in values array' curl localhost | jq '.values | last'
$ curl localhost | jq '.values | last'
- ollama - codegemma or codellama models
- python3
export AISH_TEMP_DIR="/tmp/reverse_ai_search"
export AISH_PATH="/path/to/git/repo"
export AISH_HOTKEY=${AISH_HOTKEY:-$'\C-o'}
source $AISH_PATH/autocomplete.zsh
zle -N reverse_ai_search
bindkey '^O' reverse_ai_search
bind -x '"\C-o": "reverse_ai_search"'
export AISH_TEMP_DIR="/tmp/reverse_ai_search"
export AISH_PATH="/path/to/codebase"
export AISH_HOTKEY=${AISH_HOTKEY:-'0f'}
source $AISH_PATH/autocomplete.sh
Configuration values with their defaults
# temporary directory for caching lookup
AISH_TEMP_DIR=/tmp/reverse_ai_search
# Ollama model to use
AISH_OLLAMA_MODEL=codegemma
# Ollama host string
AISH_OLLAMA_HOST=http://localhost:11434
- ZSH support
- send buffer to LLM every keystroke. Otherwise is sent on function call.
AISH_STREAM_RESULTS=1 - incorporate the directory into the prompt
- prompt tuning
- fine-tuned model
- It caches results for strings in the temp directory, so you may want to clear these for many reasons.
- There is an attempt to rate limit the number of predictions sent to ollama per session. If you are experiencing slowness disable
AISH_STREAM_RESULTSand use the key bind to initialize ai search.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent