Module
Upgrade your terminal with GPT-4. Ask questions, automate commands, pipe I/O, etc. Made with Deno.
Latest
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823import { parse } from "https://deno.land/std@0.181.0/flags/mod.ts";import { aiConfig, getChatResponse_stream, getImageResponse, StreamResponse, ChatCompletionRequest, CreateImageRequest, Message, EmbeddingRequest, getEmbeddingResponse } from "./lib/ai.ts";import { AUTO_UPDATE_PROBABILITY, Config, DEFAULT_CONFIG, getChat, getHistory, getOrCreateConfigFile, loadConfig, saveConfig, writeChat,} from "./lib/data.ts";import { printCtrlSequence, pullCharacter } from "./lib/lib.ts";import { setCodeCmdParamsForChat, setExecutableCmdParamsForChat, setLanguageForChat,} from "./lib/prompts.ts";import { exec as shExec } from "./lib/shell.ts";import { install, isLatestVersion } from "./lib/update.ts";
const args = parse(Deno.args, { boolean: [ // Instructions for this script "help",
// Runs through persistent configuration "config",
// Continuation (continue last conversation) "continue", "cont", "c",
// Exec (run as a shell command) "exec", "x",
// Retry (re-generate the last assistant message) "retry", "r",
// Rewrite (reword the last user message) "rewrite", "rw", "w",
// Pop (remove the last message in the conversation) "pop",
// Print (print the last message in the conversation) "print", "p",
// Slice (remove the first message in the conversation) "slice", "s",
// History (list chat history) "history", "h",
// Dump (dump the entire chat history) "dump", "d",
// Fast (use GPT-3.5-turbo) "fast", "f",
// Update (update ShellGPT) "update", "u",
// REPL (continuous conversation) "repl",
// Code mode (output code instead of text) "code",
// Debug (print debug info) "debug",
// System (primary input treated as system prompt) "as-system", "as-sys",
// Image "image", "img", "i",
// Embedding "embed" ], string: [ // Name (select a conversation from history to use) "name", "n",
// System (set a system prompt / context) "system", "sys",
// Temperature (creativity) "temperature", "temp", "t",
// Limit max tokens to output "max_tokens", "max",
// WPM (words per minute, speed of typing output) "wpm",
// Model (manually use a different OpenAI model) "model", "m",
// Language to use for programming or text generation 'lang',
// Dimensions for embedding "dims", ],});
// --- Parse Args ---const DEFAULT_MODEL = "gpt-4-turbo-preview";const DEFAULT_WPM = 1200;const AVG_CHARS_PER_WORD = 4.8;
const help = args.help;const name = args.name || args.n;const fast = args.f || args.fast;const updateConfig = args.config;const update = args.update || args.u;const model = fast ? "gpt-3.5-turbo-0125" as const : (args.model ?? args.m);const temp = args.t || args.temp || args.temperature;const exec = args.x || args.exec;const retry = args.r || args.retry;const rewrite = args.w || args.rw || args.rewrite;const pop = args.pop;const print = args.p || args.print;const slice = args.s || args.slice;const dump = args.dump || args.d;const cont = slice || pop || retry || rewrite || print || dump || (Boolean(args.c || args.cont || args.continue));const history = args.h || args.history;const system = args.sys || args.system;const maxTokens = args.max || args.max_tokens;const readStdin = args._.at(0) === "-" || args._.at(-1) === "-";const repl = args.repl;const code = args.code;const debug = args.debug;const asSys = args["as-sys"] || args["as-system"];const bumpSys = args["as-sys"] || args["as-system"];const lang = args.langconst img = args.img || args.iconst embed = args.embedconst dims = args.dims// --- END Parse Args ---
let config = await loadConfig();const gptCommand = config?.command ?? DEFAULT_CONFIG.command;const wpm = args.wpm ? Number(args.wpm) : config?.wpm || DEFAULT_WPM;const configWasEmpty = Object.keys(config ?? {}).length === 0;const messageWasEmpty = args._.length === 0;const shouldAutoUpdate = config?.autoUpdate !== "never" && Math.random() < AUTO_UPDATE_PROBABILITY;
const messageContent = args._.join(" ");
const message: Message = { role: asSys ? "system" : "user", content: messageContent,};
const stock: ChatCompletionRequest = { model: model ?? config?.model ?? DEFAULT_MODEL, messages: [],};
const req: ChatCompletionRequest = (cont || name) ? ((await getChat(name)) ?? stock) : stock;
const helpMessage = `Usage: ${gptCommand} [OPTIONS] [MESSAGE]
Options: --help Show this help message --config Runs configuration --update Updates ShellGPT --debug Output debug information with each request - Read message from stdin -c, --continue Continue the last conversation -x, --exec Run the generated response as a shell command -r, --retry Re-generate the last assistant message -w, --rewrite Reword the last user message -s, --slice Remove the first message in the conversation --pop Remove the last message in the conversation -p, --print Print the last message in the conversation -h, --history List chat history -d, --dump Dump the entire chat history -f, --fast Use GPT-3.5-turbo model (faster) --repl Start a continuous conversation --code Output code instead of chat text --lang LANG Set the language for programming or text generation --bump-sys Bump the most recent system prompt/context to front
-i, --img Output an image instead of text --embed Output a vector embedding of the input --dims Set the dimensions of the embedding
-n, --name NAME Select a conversation from history to use --sys[tem] Set a system prompt/context --as-sys[tem] Treat the primary input as a system prompt -t, --temp Set the creativity temperature --wpm WPM Set the words per minute --max MAX_TOKENS Set the maximum number of tokens -m, --model MODEL Manually use a different OpenAI model
Examples: ${gptCommand} "What is the capital of France?" ${gptCommand} -c "Tell me more about Paris." ${gptCommand} -x "Create a new file called 'test.txt' and write 'Hello World!' to it." cat test.txt | ${gptCommand} - "Invert the capitalization of this text."`;
// --- HANDLE ARGS ---if (debug) { aiConfig.debug = true;}if (embed) { const newReq: EmbeddingRequest = { input: messageContent, model: model ?? 'text-embedding-3-small', ...(dims ? {dimensions: Number(dims)} : {}) };
const response = await getEmbeddingResponse(newReq); if (response) { console.log(JSON.stringify(response)); } else { console.log("(Failed to generate embedding)"); } Deno.exit();}if (img) { const newReq: CreateImageRequest = { quality: "hd", response_format: "url", size: "1024x1024", prompt: messageContent ?? img, model: "dall-e-3", }; const response = await getImageResponse(newReq); if (response) { console.log(response); } else { console.log("(Failed to generate image)"); } Deno.exit();}if (pop) { const lastMessage = req.messages.pop(); if (lastMessage) { console.log(`(Removing last message from ${lastMessage!.role})`); await writeChat(req, false); } else { console.log("(Found no messages)"); } Deno.exit();}
if (slice) { if (req.messages.length > 1) { console.log(`(Removing first message)`); req.messages = req.messages.slice(1); await writeChat(req, false); } else { console.log("(Found no messages)"); } Deno.exit();}if (update || shouldAutoUpdate) { const isLatest = await isLatestVersion(); if (shouldAutoUpdate && isLatest) { // do nothing in this case } else { const newConfig = config ?? DEFAULT_CONFIG; const updateInfo = await install(newConfig, true); saveConfig(newConfig); if (updateInfo.result === "updated") { console.log("\n%c> Successfully updated!", "font-weight: bold;"); Deno.exit(); } else if (updateInfo.result === "error") { console.log( "\n%c> Encountered error while updating.", "font-weight: bold;", ); Deno.exit(); } else { console.log("\n%c> No updates found.", "font-weight: bold;"); Deno.exit(); } }}if (updateConfig || configWasEmpty) { if (configWasEmpty) { console.log("(No config found. Running initial setup...)\n"); } const newConfig: Config = { ...(config ?? DEFAULT_CONFIG), };
const updateInfo = await install(newConfig, true);
if (updateInfo.result === "updated") { console.log( "\n%c> Successfully updated! Please re-run `gpt --config`", "font-weight: bold;", ); Deno.exit(); }
const currentModel = config?.model || DEFAULT_MODEL;
console.log( "\n%c> Which OpenAI ChatGPT model would you like to use?", "font-weight: bold;", ); console.log(); const model = window.prompt( `You can enter "gpt-4" or "gpt-3.5-turbo". (Leave empty for ${currentModel}):`, );
if (model) { newConfig.model = model ?? currentModel; }
console.log( "\n%c> Would you like to set a custom system prompt to attach to each session?", "font-weight: bold;", ); if (config?.systemPrompt) { console.log(`Current system prompt: ${config.systemPrompt}`); console.log(); const newPrompt = window.prompt( `Enter new prompt (empty to keep, "clear" to remove)`, ); if (newPrompt === "clear") { newConfig.systemPrompt = undefined; } else if (newPrompt) { newConfig.systemPrompt = newPrompt; } else { newConfig.systemPrompt = config.systemPrompt; } } else { console.log(); const newPrompt = window.prompt( `Press enter to skip, or type a new prompt:`, ); if (newPrompt) { newConfig.systemPrompt = newPrompt; } }
if (!Deno.env.get("OPENAI_API_KEY")) { console.log( `\n%c> No API key detected. It's recommended to configure using the OPENAI_API_KEY environment variable, but you can also paste one here.`, "font-weight: bold;", );
console.log( "\n%c> You can get an API key here: %chttps://platform.openai.com/account/api-keys", "font-weight: bold;", "font-weight: normal;", );
if (config?.openAiApiKey) { console.log(`Current API key: ${config.openAiApiKey}`); } console.log();
const newPrompt = window.prompt( config?.openAiApiKey ? `Enter new API key (empty to keep, "clear" to remove)` : `Enter API key (empty to skip)`, ); if (newPrompt === "clear") { newConfig.openAiApiKey = undefined; } else if (newPrompt) { newConfig.openAiApiKey = String(newPrompt); } else { // do nothing } }
try { await saveConfig(newConfig); console.log( `%cUpdated config file: %c${await getOrCreateConfigFile()}`, "color: green; font-weight: bold;", "color: green;", );
if (updateConfig) { // Exit only if the user manually requested config update Deno.exit(); } else { // Otherwise, continue with the rest of the script config = newConfig; req.model = model ?? config.model ?? DEFAULT_MODEL; } } catch (e) { console.error( `Failed to update config file at: ${await getOrCreateConfigFile()}`, ); console.error(e); Deno.exit(1); }}
if (help) { console.log(helpMessage); Deno.exit();}if (readStdin) { message.content += "\n";
const decoder = new TextDecoder(); for await (const chunk of Deno.stdin.readable) { const textChunk = decoder.decode(chunk); message.content += textChunk; }}
if (print) { // print last message const lastMessage = req.messages.pop(); if (lastMessage) { console.log(lastMessage.content); } else { console.log("(Found no messages)"); } Deno.exit();}
if (dump) { for (const message of req.messages) { if (message.role === "user") { console.log("---\n"); console.log(`${message.content}\n`); console.log("---\n"); } else { console.log(`${message.content}\n`); } } Deno.exit();}
if (system || (config?.systemPrompt && !cont)) { // Add system prompt if set for this message, or is a new conversation req.messages.push({ role: "system", content: system ?? config!.systemPrompt!, });}
let empty = false;if (!message.content && !retry && !pop && !history) { empty = true; console.log("(No message passed)\n");}
if (temp) { req.temperature = Number(temp);}
if (maxTokens) { req.max_tokens = Number(maxTokens);}
if (model) { // @ts-ignore Allow any string as model for now req.model = model;}
if (retry) { // remove last assistant message req.messages.pop();}if (rewrite) { // remove last assistant AND user messages req.messages.pop(); req.messages.pop();}if (!retry && !empty) { req.messages.push(message);}
if (exec) { setExecutableCmdParamsForChat(req);} else if (code) { setCodeCmdParamsForChat(req, lang);} else if (lang) { setLanguageForChat(req, lang)}
if (history) { const files = await getHistory(); for (const file of files) { const hasSnippet = file.snippet && file.snippet.length > 0; if (hasSnippet) { console.log( `%c${file.name}\t\t%c${file.snippet}`, "color: blue", "color: gray", ); } else { console.log(`%c${file.name}`, "color: blue"); } } Deno.exit();}
if (bumpSys) { let mostRecentSystemIndex = -1; for (let i = req.messages.length - 1; i > 0; i--) { if (req.messages[i].role === "system") { mostRecentSystemIndex = i; break; } }
// splice+push it if (mostRecentSystemIndex !== -1) { const systemMessage = req.messages.splice(mostRecentSystemIndex, 1)[0]; req.messages.push(systemMessage); }}// --- END HANDLE ARGS ---
let streamResponse: AsyncIterableIterator<StreamResponse> | null = null;
const doStreamResponse = async () => { try { streamResponse = await getChatResponse_stream(req); } catch (e) { console.error("Unhandled error", e); Deno.exit(); }};
// STATEtype DoneType = "none" | "with_net" | "with_write" | "with_print";let done: DoneType = "none"; // none -> with_net -> with_write -> with_print -> [done]let responseStr = "";let intermediateStr = "";let printStr = "";
if (repl && messageWasEmpty) { done = "with_net";} else { await doStreamResponse();}
let nextBuffer = "";async function readInput() { let strBuffer = ""; let arrayBuffer = new Uint8Array(1); // Buffer to hold byte data from stdin
let done = false; let newline = false;
const updateFn = async () => { // read from stdin const n = await Deno.stdin.read(arrayBuffer)
let curBuffer = "" const wasDone = done
if (n != null && n > 0) { const text = new TextDecoder().decode(arrayBuffer)
for (const char of text) { const code = char.charCodeAt(0); if (code === 13 || code === 10) { // carriage return or newline newline = true curBuffer += "\n"; await Deno.stdout.write(new TextEncoder().encode("\n")); } else if (code === 127) { // backspace const wasEmpty = strBuffer.length === 0; curBuffer = curBuffer.slice(0, -1); strBuffer = strBuffer.slice(0, -1);
if (!wasEmpty) { await Deno.stdout.write(new TextEncoder().encode("\b \b")); } } else { // any other character // console.warn("Z", char) await Deno.stdout.write(arrayBuffer); curBuffer += char; newline = false; } }
if (wasDone) { // We already finished reading this stream, so push to next reader nextBuffer = curBuffer } else { strBuffer += curBuffer }
} else { done = true; newline = true; return; } };
nextBuffer = "" Deno.stdin.setRaw(false) Deno.stdin.setRaw(true, { cbreak: true })
await new Promise((resolve) => { ( async () => { strBuffer = "" arrayBuffer = new Uint8Array(1) while (!done) { if (nextBuffer) { // HACK: there may be another stdin reader which we want to push results from // Use this instead of AbortController for now strBuffer = `${nextBuffer}${strBuffer}` nextBuffer = "" } const wasNewline = newline; if (wasNewline) { // wait for a moment for typing setTimeout(() => { if (!newline) { return } done = true resolve(true) }, 250) } await updateFn(); if (wasNewline && newline) { done = true; } } resolve(true) } )() })
Deno.stdin.setRaw(false)
return strBuffer}
// Done, write it outconst flush = async () => { streamResponse = null; const text = new TextEncoder().encode("\n"); await Deno.stdout.write(text);
req.messages.push({ content: responseStr, role: "assistant", });
await writeChat(req, cont ? false : (name || true));
if (exec && !readStdin) { console.log( `\n%cAre you SURE you wish to run the above command? (y/N):`, "color: red; font-weight: bold;", ); const promptValue = prompt("> "); if (["y", "yes"].includes(promptValue?.toLowerCase() ?? "")) { // do it await shExec(responseStr); } else { console.log("(will not exec command)"); } } else if (exec && readStdin) { console.log("(exec not currently supported when reading from stdin)"); }
if (repl) { responseStr = "";
await printCtrlSequence("blue"); await printCtrlSequence("bold");
await Deno.stdout.write(new TextEncoder().encode(`\n> `));
const promptValue = await readInput();
// print reset ctrl sequence await printCtrlSequence("reset");
if (promptValue) { // do it req.messages.push({ content: promptValue, role: "user", }); done = "none"; await doStreamResponse(); } } else { Deno.exit(); }};
// Push strings{ (async () => { while (true) { if (done !== "none" || !streamResponse) { // Spin wait await new Promise((resolve) => setTimeout(resolve)); continue; } try { const iterator = streamResponse as AsyncIterableIterator<StreamResponse>;
while (true) { const response = await iterator.next(); if (response.value.delta) { responseStr += response.value.delta;
if (wpm >= 0) { // Write to a buffer intermediateStr += response.value.delta; } else { // Print immediately printStr += response.value.delta; } } if (response.done) { break; } } } catch (e) { console.error("Unhandled error", e); Deno.exit(1); }
if (wpm >= 0) { done = "with_net"; } else { done = "with_write"; } } })();}
if (wpm >= 0) { // Writes to a buffer so we can print at a desired WPM let startTime = -1; const targetCps = (AVG_CHARS_PER_WORD * wpm) / 60; (async () => { while (true) { if (done === "with_write" || (done as DoneType) === "with_print") { // Spin wait await new Promise((resolve) => setTimeout(resolve)); continue; } // Go through characters one-by-one and write while ((done as DoneType) !== "with_net" || intermediateStr.length > 0) { if (startTime < 0) { startTime = Date.now(); } const { char, str } = pullCharacter(intermediateStr);
printStr += char; intermediateStr = str; await new Promise((resolve) => { setTimeout(() => { resolve(true); }, 1000 / targetCps); }); } done = "with_write"; } })();}
// Pull strings to write as soon as available{ const consumeFn = async () => { const latest = printStr; printStr = ""; if (!latest && done === "with_write") { await flush(); } if (latest) { const text = new TextEncoder().encode(latest); await Deno.stdout.write(text); } setTimeout(consumeFn); }; setTimeout(consumeFn);}