├── LICENSE ├── README.md └── openai /LICENSE: -------------------------------------------------------------------------------- 1 | MIT License 2 | 3 | Copyright (c) 2023 Janlay Wu 4 | 5 | Permission is hereby granted, free of charge, to any person obtaining a copy 6 | of this software and associated documentation files (the "Software"), to deal 7 | in the Software without restriction, including without limitation the rights 8 | to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 9 | copies of the Software, and to permit persons to whom the Software is 10 | furnished to do so, subject to the following conditions: 11 | 12 | The above copyright notice and this permission notice shall be included in all 13 | copies or substantial portions of the Software. 14 | 15 | THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 16 | IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 17 | FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 18 | AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 19 | LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 20 | OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 21 | SOFTWARE. 22 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | # openai-cli 2 | A universal cli for OpenAI, written in BASH. 3 | 4 | # Features 5 | - [x] Scalable architecture allows for continuous support of new APIs. 6 | - [x] Custom API name, version, and all relevant properties. 7 | - [x] Dry-run mode (without actually initiating API calls) to facilitate debugging of APIs and save costs. 8 | - [x] New in v3: Supports any AI services that provide OpenAI-compatible APIs. 9 | 10 | Important changes in version 3: 11 | - `-v api_version` in previous versions is now removed. If you have a custom `OPENAI_API_ENDPOINT`, you need to append API version in it. The internal API version is remove as some services like DeepSeek don't have version prefix in `/v1/chat/completions`. 12 | - `OPENAI_CHAT_MODEL` no longer supported, use `OPENAI_API_MODEL` instead. 13 | - By default, the request no longer includes optional parameters `temperature` / `max_tokens`. You need to explictly add `+temperature` / `+max_tokens` to customize if necessary. 14 | 15 | Available APIs: 16 | - [x] `chat/completions` (default API) 17 | - [x] `models` 18 | - [x] `images/generations` 19 | - [x] `embeddings` 20 | - [x] `moderations` 21 | 22 | The default API `chat/completions` provides: 23 | - [x] Complete pipelining to interoperate with other applications 24 | - [x] Allow prompts to be read from command line arguments, file, and stdin 25 | - [x] Support streaming 26 | - [x] Support multiple topics 27 | - [x] Support continuous conversations. 28 | - [ ] Token usage 29 | 30 | # Installation 31 | - [jq](https://stedolan.github.io/jq/) is required. 32 | - Linux: `sudo apt install jq` 33 | - macOS: `brew install jq` 34 | - Download script and mark it executable: 35 | ```bash 36 | curl -fsSLOJ https://go.janlay.com/openai 37 | chmod +x openai 38 | ``` 39 | You may want to add this file to a directory in `$PATH`. 40 | 41 | Also install the manual page, e.g.: 42 | ```bash 43 | pandoc -s -f markdown -t man README.md > /usr/local/man/man1/openai.1 44 | ``` 45 |

46 |

Further reading: curl's killer feature

47 | -OJ is a killer feature 48 |

49 | 50 | Now you can try it out! 51 | 52 | # Tips 53 | ## Getting started 54 | To begin, type `openai -h` to access the help manual. 55 | 56 | ⚠️ If you run `openai` directly, it may appear to be stuck because it expects prompt content from stdin which is not yet available. To exit, simply press Ctrl+C to interrupt the process. 57 | 58 |

59 |

Why are you so serious?

60 | 61 | What happens when the `openai` command is executed without any parameters? It means that: 62 | - The default API used will be `chat/completions`, and the schema version will be `v1`. 63 | - The prompt will be read from stdin. 64 | - The program will wait for input while stdin remains empty. 65 |

147 |

Command and output

148 | 149 | ``` 150 | $ openai -n hello, world! 151 | Dry-run mode, no API calls made. 152 | 153 | Request URL: 154 | -------------- 155 | https://api.openai.com/v1/chat/completions 156 | 157 | Authorization: 158 | -------------- 159 | Bearer sk-cfw****NYre 160 | 161 | Payload: 162 | -------------- 163 | { 164 | "model": "gpt-3.5-turbo", 165 | "temperature": 0.5, 166 | "max_tokens": 200, 167 | "stream": true, 168 | "messages": [ 169 | { 170 | "role": "user", 171 | "content": "hello, world!" 172 | } 173 | ] 174 | } 175 | ``` 176 |

177 | With full pipelining support, you can achieve the same functionality using alternative methods: 178 | 179 | ```bash 180 | echo 'hello, world!' | openai -n 181 | ``` 182 | 183 |

184 |

For BASH gurus

185 | 186 | This would be same: 187 | ```bash 188 | echo 'hello, world!' >hello.txt 189 | openai -n 202 | 203 | It seems you have understood the basic usage. Try to get real answer from OpenAI: 204 | 205 | ```bash 206 | openai hello, world! 207 | ``` 208 | 209 |

210 |

Command and output

211 | 212 | ``` 213 | $ openai hello, world! 214 | Hello there! How can I assist you today? 215 | ``` 216 | 217 |

218 | 219 | ## Topics 220 | Topic starts with a `@` sign. so `openai @translate Hello, world!` means calling the specified topic `translate`. 221 | 222 | To create new topic, like translate, with the initial prompt (system role, internally): 223 | ```bash 224 | openai @translate 'Translate, no other words: Chinese -> English, Non-Chinese -> Chinese' 225 | ``` 226 | 227 | Then you can use the topic by 228 | ```bash 229 | openai @translate 'Hello, world!' 230 | ``` 231 | You should get answer like `你好，世界！`. 232 | 233 | Again, to see what happens, use the dry-run mode by adding `-n`. You will see the payload would be sent: 234 | ```json 235 | { 236 | "model": "gpt-3.5-turbo", 237 | "temperature": 0.5, 238 | "max_tokens": 200, 239 | "stream": true, 240 | "messages": [ 241 | { 242 | "role": "system", 243 | "content": "Translate, no other words: Chinese -> English, Non-Chinese -> Chinese" 244 | }, 245 | { 246 | "role": "user", 247 | "content": "Hello, world!" 248 | } 249 | ] 250 | } 251 | ``` 252 | 253 | ## Chatting 254 | All use cases above are standalone queries, not converstaions. To chat with OpenAI, use `-c`. This can also continue existing topic conversation by prepending `@topic`. 255 | 256 | Please note that chat requests will quickly consume tokens, leading to increased costs. 257 | 258 | ## Advanced 259 | To be continued. 260 | 261 | ## Manual 262 | To be continued. 263 | 264 | # LICENSE 265 | This project uses the MIT license. Please see [LICENSE](https://github.com/janlay/openai-cli/blob/master/LICENSE) for more information. 266 | -------------------------------------------------------------------------------- /openai: -------------------------------------------------------------------------------- 1 | #!/usr/bin/env bash 2 | # 3 | # OpenAI CLI v3.0 4 | # Created by @janlay 5 | # 6 | 7 | set -eo pipefail 8 | 9 | declare _config_dir="${OPENAI_DATA_DIR:-$XDG_CONFIG_HOME}" 10 | OPENAI_DATA_DIR="${_config_dir:-$HOME/.openai}" 11 | 12 | # Read config file? 13 | [ -e "$OPENAI_DATA_DIR"/config ] && . "$OPENAI_DATA_DIR"/config 14 | 15 | # openai-cli accepts various exported environment variables: 16 | # OPENAI_API_KEY : OpenAI's API key 17 | # OPENAI_API_ENDPOINT : Custom API endpoint 18 | # OPENAI_API_MODEL : Which model to use 19 | # OPENAI_DATA_DIR : Directory to store data 20 | OPENAI_API_ENDPOINT="${OPENAI_API_ENDPOINT:-https://api.openai.com/v1}" 21 | OPENAI_API_KEY="${OPENAI_API_KEY:-}" 22 | OPENAI_API_MODEL="${OPENAI_API_MODEL:-gpt-4o}" 23 | 24 | # defaults 25 | readonly _app_name=openai _app_version=3.0 26 | readonly provider="${OPENAI_COMPATIBLE_PROVIDER:-OPENAI}" 27 | readonly default_api_name=chat/completions default_model="$OPENAI_API_MODEL" default_topic=General 28 | 29 | declare -i chat_mode=0 dry_run=0 30 | declare tokens_file="$OPENAI_DATA_DIR/total_tokens" api_name=$default_api_name topic=$default_topic 31 | declare dump_file dumped_file data_file temp_dir rest_args prompt_file prompt 32 | 33 | trap cleanup EXIT 34 | cleanup() { 35 | if [ -d "$temp_dir" ]; then 36 | rm -rf -- "$temp_dir" 37 | fi 38 | } 39 | 40 | get_env() { 41 | local env_key="${provider}_${1}" 42 | echo "${!env_key}" 43 | } 44 | 45 | raise_error() { 46 | [ "$2" = 0 ] || echo -n "$_app_name: " >&2 47 | echo -e "$1" >&2 48 | exit "${2:-1}" 49 | } 50 | 51 | load_conversation() { 52 | [ -f "$data_file" ] && cat "$data_file" || echo '{}' 53 | } 54 | 55 | update_conversation() { 56 | local entry="$2" data 57 | [[ $entry == \{* ]] || entry=$(jq -n --arg content "$entry" '{$content}') 58 | entry=$(jq --arg role "$1" '. += {$role}' <<<"$entry") 59 | data=$(load_conversation) 60 | jq --argjson item "$entry" '.messages += [$item]' <<<"$data" >"$data_file" 61 | } 62 | 63 | save_tokens() { 64 | local data num="$1" 65 | [ -f "$data_file" ] && { 66 | data=$(load_conversation) 67 | jq --argjson tokens "$num" '.total_tokens += $tokens' <<<"$data" >"$data_file" 68 | } 69 | 70 | data=0 71 | [ -f "$tokens_file" ] && data=$(cat "$tokens_file") 72 | echo "$((data + num))" >"$tokens_file" 73 | } 74 | 75 | read_prompt() { 76 | # read prompt from args first 77 | local word accepts_props=1 props='{}' real_prompt 78 | if [ ${#rest_args[@]} -gt 0 ]; then 79 | # read file $prompt_file word by word, and extract words starting with '+' 80 | for word in "${rest_args[@]}"; do 81 | if [ $accepts_props -eq 1 ] && [ "${word:0:1}" = '+' ]; then 82 | word="${word:1}" 83 | # determine value's type for jq 84 | local options=(--arg key "${word%%=*}") value="${word#*=}" arg=--arg 85 | [[ $value =~ ^[+-]?\ ?[0-9.]+$ || $value = true || $value = false || $value == [\[\{]* ]] && arg=--argjson 86 | options+=("$arg" value "$value") 87 | props=$(jq "${options[@]}" '.[$key] = $value' <<<"$props") 88 | else 89 | real_prompt="$real_prompt $word" 90 | accepts_props=0 91 | fi 92 | done 93 | [ -n "$props" ] && echo "$props" >"$temp_dir/props" 94 | fi 95 | 96 | if [ -n "$real_prompt" ]; then 97 | [ -n "$prompt_file" ] && echo "* Prompt file \`$prompt_file' will be ignored as the prompt parameters are provided." >&2 98 | echo -n "${real_prompt:1}" >"$temp_dir/prompt" 99 | elif [ -n "$prompt_file" ]; then 100 | [ -f "$prompt_file" ] || raise_error "File not found: $prompt_file." 3 101 | [[ -s $prompt_file ]] || raise_error "Empty file: $prompt_file." 4 102 | fi 103 | } 104 | 105 | openai_models() { 106 | call_api | jq 107 | } 108 | 109 | openai_moderations() { 110 | local prop_file="$temp_dir/props" payload="{\"model\": \"text-moderation-latest\"}" 111 | 112 | # overwrite default properties with user's 113 | read_prompt 114 | [ -f "$prop_file" ] && payload=$(jq -n --argjson payload "$payload" '$payload | . += input' <"$prop_file") 115 | 116 | # append user's prompt to messages 117 | local payload_file="$temp_dir/payload" input_file="$temp_dir/prompt" 118 | [ -f "$input_file" ] || input_file="${prompt_file:-/dev/stdin}" 119 | jq -Rs -cn --argjson payload "$payload" '$payload | .input = input' "$input_file" >"$payload_file" 120 | 121 | call_api | jq -c '.results[]' 122 | } 123 | 124 | openai_images_generations() { 125 | local prop_file="$temp_dir/props" payload="{\"n\": 1, \"size\": \"1024x1024\"}" 126 | 127 | # overwrite default properties with user's 128 | read_prompt 129 | [ -f "$prop_file" ] && payload=$(jq -n --argjson payload "$payload" '$payload | . += input | . += {response_format: "url"}' <"$prop_file") 130 | 131 | # append user's prompt to messages 132 | local payload_file="$temp_dir/payload" input_file="$temp_dir/prompt" 133 | [ -f "$input_file" ] || input_file="${prompt_file:-/dev/stdin}" 134 | jq -Rs -cn --argjson payload "$payload" '$payload | .prompt = input' "$input_file" >"$payload_file" 135 | 136 | call_api | jq -r '.data[].url' 137 | } 138 | 139 | openai_embeddings() { 140 | local prop_file="$temp_dir/props" payload="{\"model\": \"text-embedding-ada-002\"}" 141 | 142 | # overwrite default properties with user's 143 | read_prompt 144 | [ -f "$prop_file" ] && payload=$(jq -n --argjson payload "$payload" '$payload | . += input' <"$prop_file") 145 | 146 | # append user's prompt to messages 147 | local payload_file="$temp_dir/payload" input_file="$temp_dir/prompt" 148 | [ -f "$input_file" ] || input_file="${prompt_file:-/dev/stdin}" 149 | jq -Rs -cn --argjson payload "$payload" '$payload | .input = input' "$input_file" >"$payload_file" 150 | 151 | call_api | jq -c 152 | } 153 | 154 | openai_chat_completions() { 155 | local streaming=0 156 | if [ -n "$dumped_file" ]; then 157 | # only succeeds when the dumped file is not streamed 158 | jq -er '.choices[0].message.content' <"$dumped_file" 2>/dev/null && return 159 | streaming=1 160 | else 161 | local prop_file="$temp_dir/props" model payload 162 | model=$(get_env API_MODEL) 163 | payload="{\"model\": \"$model\", \"stream\": true}" 164 | 165 | # overwrite default properties with user's 166 | read_prompt 167 | [ -f "$prop_file" ] && { 168 | payload=$(jq -n --argjson payload "$payload" '$payload | . += input | . += {messages: []}' <"$prop_file") 169 | } 170 | 171 | local data 172 | data=$(load_conversation | jq .messages) 173 | [ "$topic" != "$default_topic" ] && { 174 | if [ $chat_mode -eq 1 ]; then 175 | # load all messages for chat mode 176 | payload=$(jq --argjson messages "$data" 'setpath(["messages"]; $messages)' <<<"$payload") 177 | else 178 | # load only first message for non-chat mode 179 | payload=$(jq --argjson messages "$data" 'setpath(["messages"]; [$messages[0]])' <<<"$payload") 180 | fi 181 | } 182 | # append user's prompt to messages 183 | local payload_file="$temp_dir/payload" input_file="$temp_dir/prompt" 184 | [ -f "$input_file" ] || input_file="${prompt_file:-/dev/stdin}" 185 | jq -Rs -cn --argjson payload "$payload" '$payload | .messages += [{role: "user", content: input}]' "$input_file" >"$payload_file" 186 | 187 | streaming=$(jq -e 'if .stream then 1 else 0 end' <"$payload_file") 188 | 189 | # check o1's parameters 190 | jq -e 'select(.model | test("^o\\d")) and (.temperature or .top_p or .presence_penalty or .frequency_penalty .logprobs or .top_logprobs .logit_bias)' <"$payload_file" &>/dev/null && raise_error 'One or more unsupported API parameters used for model o1. See https://platform.openai.com/docs/guides/reasoning#limitations for more details.' 5 191 | fi 192 | 193 | local chunk reason text role fn_name 194 | if [ $streaming -eq 1 ]; then 195 | call_api | while read -r chunk; do 196 | [ -z "$chunk" ] && continue 197 | chunk=$(cut -d: -f2- <<<"$chunk" | jq '.choices[0]') 198 | reason=$(jq -r '.finish_reason // empty' <<<"$chunk") 199 | [[ $reason = stop || $reason = function_call ]] && break 200 | [ -n "$reason" ] && raise_error "API error: $reason" 10 201 | 202 | # get role and function info from the first chunk 203 | [ -z "$role" ] && { 204 | role=$(jq -r '.delta.role // empty' <<<"$chunk") 205 | fn_name=$(jq -r '.delta.function_call.name // empty' <<<"$chunk") 206 | } 207 | 208 | # workaround: https://stackoverflow.com/a/15184414 209 | chunk=$( 210 | jq -r '.delta | .function_call.arguments // .content // empty' <<<"$chunk" 211 | printf x 212 | ) 213 | # ensure chunk is not empty 214 | [ ${#chunk} -ge 2 ] || continue 215 | 216 | chunk="${chunk:0:${#chunk}-2}" 217 | text="$text$chunk" 218 | echo -n "$chunk" 219 | done 220 | [ "$dry_run" -eq 0 ] && echo 221 | else 222 | text=$(call_api | jq -er '.choices[0].message.content') 223 | echo "$text" 224 | fi 225 | 226 | # append response to topic file for chat mode 227 | if [ "$chat_mode" -eq 1 ]; then 228 | [ -n "$fn_name" ] && text=$(jq -n --arg name "$fn_name" --argjson arguments "${text:-\{\}}" '{function_call: {$name, $arguments}}') 229 | 230 | update_conversation user "$prompt" 231 | update_conversation "$role" "$text" 232 | fi 233 | } 234 | 235 | # shellcheck disable=SC2120 236 | call_api() { 237 | # return dumped file if specified 238 | [ -n "$dumped_file" ] && { 239 | cat "$dumped_file" 240 | return 241 | } 242 | 243 | local url="$(get_env API_ENDPOINT)/$api_name" auth="Bearer $(get_env API_KEY)" 244 | 245 | # dry-run mode 246 | [ "$dry_run" -eq 1 ] && { 247 | echo "Dry-run mode, no API calls made." 248 | echo -e "\nRequest URL:\n--------------\n$url" 249 | echo -en "\nAuthorization:\n--------------\n" 250 | sed -E 's/(sk-.{3}).{41}/\1****/' <<<"$auth" 251 | [ -n "$payload_file" ] && { 252 | echo -e "\nPayload:\n--------------" 253 | jq <"$payload_file" 254 | } 255 | exit 0 256 | } >&2 257 | 258 | local args=("$url" --no-buffer -fsSL -H 'Content-Type: application/json' -H "Authorization: $auth") 259 | [ -n "$payload_file" ] && args+=(-d @"$payload_file") 260 | [ $# -gt 0 ] && args+=("$@") 261 | 262 | [ -n "$dump_file" ] && args+=(-o "$dump_file") 263 | curl "${args[@]}" 264 | [ -z "$dump_file" ] || exit 0 265 | } 266 | 267 | create_topic() { 268 | update_conversation system "${rest_args[*]}" 269 | raise_error "Topic '$topic' created with initial prompt '${rest_args[*]}'" 0 270 | } 271 | 272 | usage() { 273 | raise_error "OpenAI Client v$_app_version 274 | 275 | SYNOPSIS 276 | ABSTRACT 277 | $_app_name [-n] [-a api_name] [-o dump_file] [INPUT...] 278 | $_app_name -i dumped_file 279 | 280 | DEFAULT_API ($default_api_name) 281 | $_app_name [-c] [+property=value...] [@TOPIC] [-f file | prompt ...] 282 | prompt 283 | Prompt string for the request to OpenAI API. This can consist of multiple 284 | arguments, which are considered to be separated by spaces. 285 | -f file 286 | A file to be read as prompt. If file is - or neither this parameter nor a prompt 287 | is specified, read from standard input. 288 | -c 289 | Continues the topic, the default topic is '$default_topic'. 290 | property=value 291 | Overwrites default properties in payload. Prepend a plus sign '+' before property=value. 292 | eg: +model=gpt-3.5-turbo-0301, +stream=false 293 | 294 | TOPICS 295 | Topic starts with an at sign '@'. 296 | To create new topic, use \`$_app_name @new_topic initial prompt' 297 | 298 | OTHER APIS 299 | $_app_name -a models 300 | 301 | GLOBAL OPTIONS 302 | Global options apply to all APIs. 303 | -a name 304 | API name, default is '$default_api_name'. 305 | -n 306 | Dry-run mode, don't call API. 307 | -o filename 308 | Dumps API response to a file and exits. 309 | -i filename 310 | Uses specified dumped file instead of requesting API. 311 | Any request-related arguments and user input are ignored. 312 | 313 | -- 314 | Ignores rest of arguments, useful when unquoted prompt consists of '-'. 315 | 316 | -h 317 | Shows this help" 0 318 | } 319 | 320 | parse() { 321 | local opt 322 | while getopts 'v:a:f:i:o:cnh' opt; do 323 | case "$opt" in 324 | c) 325 | chat_mode=1 326 | ;; 327 | a) 328 | api_name="$OPTARG" 329 | ;; 330 | f) 331 | prompt_file="$OPTARG" 332 | [ "$prompt_file" = - ] && prompt_file= 333 | ;; 334 | n) 335 | dry_run=1 336 | ;; 337 | i) 338 | dumped_file="$OPTARG" 339 | ;; 340 | o) 341 | dump_file="$OPTARG" 342 | ;; 343 | h | ?) 344 | usage 345 | ;; 346 | esac 347 | done 348 | shift "$((OPTIND - 1))" 349 | 350 | # extract the leading topic 351 | [[ "$1" =~ ^@ ]] && { 352 | topic="${1#@}" 353 | shift 354 | } 355 | 356 | [ $chat_mode -eq 0 ] || { 357 | [[ -n $topic && $topic != "$default_topic" ]] || raise_error 'Topic is required for chatting.' 2 358 | } 359 | 360 | rest_args=("$@") 361 | } 362 | 363 | check_bin() { 364 | command -v "$1" >/dev/null || raise_error "$1 not found. Use package manager (Homebrew, apt-get etc.) to install it." "${2:-1}" 365 | } 366 | 367 | main() { 368 | # show compatible provider if applicable 369 | if [ "${SUPPRESS_PROVIDER_TIPS:-0}" -eq 0 ] && [ -n "$OPENAI_COMPATIBLE_PROVIDER" ]; then 370 | echo "OpenAI compatible provider: $OPENAI_COMPATIBLE_PROVIDER" >&2 371 | fi 372 | 373 | # check required config 374 | local key 375 | for key in API_ENDPOINT API_KEY API_MODEL; do 376 | [ -z "$(get_env "$key")" ] && raise_error "Missing environment variable: ${provider}_$key." 1 377 | done 378 | 379 | parse "$@" 380 | check_bin jq 10 381 | 382 | mkdir -p "$OPENAI_DATA_DIR" 383 | data_file="$OPENAI_DATA_DIR/$topic.json" 384 | temp_dir=$(mktemp -d) 385 | 386 | if [[ $topic == "$default_topic" || -f "$data_file" ]]; then 387 | local fn="openai_${api_name//\//_}" 388 | [ "$(type -t "$fn")" = function ] || raise_error "API '$api_name' is not available." 12 389 | "$fn" 390 | else 391 | [ ${#rest_args[@]} -gt 0 ] || raise_error "Prompt for new topic is required" 13 392 | create_topic 393 | fi 394 | } 395 | 396 | main "$@" 397 | --------------------------------------------------------------------------------