1<?php
2
3use dokuwiki\Extension\CLIPlugin;
4use dokuwiki\plugin\aichat\AbstractCLI;
5use dokuwiki\plugin\aichat\Chunk;
6use dokuwiki\plugin\aichat\ModelFactory;
7use dokuwiki\Search\Indexer;
8use splitbrain\phpcli\Colors;
9use splitbrain\phpcli\Options;
10use splitbrain\phpcli\TableFormatter;
11
12/**
13 * DokuWiki Plugin aichat (CLI Component)
14 *
15 * @license GPL 2 http://www.gnu.org/licenses/gpl-2.0.html
16 * @author  Andreas Gohr <gohr@cosmocode.de>
17 */
18class cli_plugin_aichat extends AbstractCLI
19{
20    /** @var helper_plugin_aichat */
21    protected $helper;
22
23    /** @inheritDoc */
24    protected function setup(Options $options)
25    {
26        parent::setup($options);
27
28        $options->setHelp(
29            'Manage and query the AI chatbot data. Please note that calls to your LLM provider will be made. ' .
30            'This may incur costs.'
31        );
32
33        $options->registerOption(
34            'model',
35            'Overrides the chat and rephrasing model settings and uses this model instead',
36            '',
37            'model'
38        );
39
40        $options->registerCommand(
41            'embed',
42            'Create embeddings for all pages. This skips pages that already have embeddings'
43        );
44        $options->registerOption(
45            'clear',
46            'Clear all existing embeddings before creating new ones',
47            'c',
48            false,
49            'embed'
50        );
51
52        $options->registerCommand('maintenance', 'Run storage maintenance. Refer to the documentation for details.');
53
54        $options->registerCommand('similar', 'Search for similar pages');
55        $options->registerArgument('query', 'Look up chunks similar to this query', true, 'similar');
56
57        $options->registerCommand('ask', 'Ask a question');
58        $options->registerArgument('question', 'The question to ask', true, 'ask');
59
60        $options->registerCommand('chat', 'Start an interactive chat session');
61
62        $options->registerCommand('models', 'List available models');
63
64        $options->registerCommand('info', 'Get Info about the vector storage and other stats');
65
66        $options->registerCommand('split', 'Split a page into chunks (for debugging)');
67        $options->registerArgument('page', 'The page to split', true, 'split');
68
69        $options->registerCommand('page', 'Check if chunks for a given page are available (for debugging)');
70        $options->registerArgument('page', 'The page to check', true, 'page');
71        $options->registerOption('dump', 'Dump the chunks', 'd', false, 'page');
72
73        $options->registerCommand('tsv', 'Create TSV files for visualizing at http://projector.tensorflow.org/' .
74            ' Not supported on all storages.');
75        $options->registerArgument('vector.tsv', 'The vector file', false, 'tsv');
76        $options->registerArgument('meta.tsv', 'The meta file', false, 'tsv');
77    }
78
79    /** @inheritDoc */
80    protected function main(Options $options)
81    {
82        parent::main($options);
83
84        $model = $options->getOpt('model');
85        if ($model) {
86            $this->helper->updateConfig(
87                ['chatmodel' => $model, 'rephasemodel' => $model]
88            );
89        }
90
91        switch ($options->getCmd()) {
92            case 'embed':
93                $this->createEmbeddings($options->getOpt('clear'));
94                break;
95            case 'maintenance':
96                $this->runMaintenance();
97                break;
98            case 'similar':
99                $this->similar($options->getArgs()[0]);
100                break;
101            case 'ask':
102                $this->ask($options->getArgs()[0]);
103                break;
104            case 'chat':
105                $this->chat();
106                break;
107            case 'models':
108                $this->models();
109                break;
110            case 'split':
111                $this->split($options->getArgs()[0]);
112                break;
113            case 'page':
114                $this->page($options->getArgs()[0], $options->getOpt('dump'));
115                break;
116            case 'info':
117                $this->showinfo();
118                break;
119            case 'tsv':
120                $args = $options->getArgs();
121                $vector = $args[0] ?? 'vector.tsv';
122                $meta = $args[1] ?? 'meta.tsv';
123                $this->tsv($vector, $meta);
124                break;
125            default:
126                echo $options->help();
127        }
128    }
129
130    /**
131     * @return void
132     */
133    protected function showinfo()
134    {
135        $stats = [
136            'embed model' => (string) $this->helper->getEmbeddingModel(),
137            'rephrase model' => (string) $this->helper->getRephraseModel(),
138            'chat model' => (string) $this->helper->getChatModel(),
139        ];
140        $stats = array_merge(
141            $stats,
142            $this->helper->getRunData(),
143            $this->helper->getStorage()->statistics()
144        );
145        $this->printTable($stats);
146    }
147
148    /**
149     * Print key value data as tabular data
150     *
151     * @param array $data
152     * @param int $level
153     * @return void
154     */
155    protected function printTable($data, $level = 0)
156    {
157        $tf = new TableFormatter($this->colors);
158        foreach ($data as $key => $value) {
159            if (is_array($value)) {
160                echo $tf->format(
161                    [$level * 2, 20, '*'],
162                    ['', $key, ''],
163                    [Colors::C_LIGHTBLUE, Colors::C_LIGHTBLUE, Colors::C_LIGHTBLUE]
164                );
165                $this->printTable($value, $level + 1);
166            } else {
167                echo $tf->format(
168                    [$level * 2, 20, '*'],
169                    ['', $key, $value],
170                    [Colors::C_LIGHTBLUE, Colors::C_LIGHTBLUE, Colors::C_LIGHTGRAY]
171                );
172            }
173        }
174    }
175
176    /**
177     * Check chunk availability for a given page
178     *
179     * @param string $page
180     * @return void
181     */
182    protected function page($page, $dump = false)
183    {
184        $indexer = new Indexer();
185        $pages = $indexer->getPages();
186        $pos = array_search(cleanID($page), $pages);
187
188        if ($pos === false) {
189            $this->error('Page not found');
190            return;
191        }
192
193        $storage = $this->helper->getStorage();
194        $chunks = $storage->getPageChunks($page, $pos * 100);
195        if ($chunks) {
196            $this->success('Found ' . count($chunks) . ' chunks');
197            if ($dump) {
198                echo json_encode($chunks, JSON_PRETTY_PRINT);
199            }
200        } else {
201            $this->error('No chunks found');
202        }
203    }
204
205    /**
206     * Split the given page into chunks and print them
207     *
208     * @param string $page
209     * @return void
210     * @throws Exception
211     */
212    protected function split($page)
213    {
214        $chunks = $this->helper->getEmbeddings()->createPageChunks($page, 0);
215        foreach ($chunks as $chunk) {
216            echo $chunk->getText();
217            echo "\n";
218            $this->colors->ptln('--------------------------------', Colors::C_LIGHTPURPLE);
219        }
220        $this->success('Split into ' . count($chunks) . ' chunks');
221    }
222
223    /**
224     * Interactive Chat Session
225     *
226     * @return void
227     * @throws Exception
228     */
229    protected function chat()
230    {
231        $history = [];
232        while ($q = $this->readLine('Your Question')) {
233            $this->helper->getChatModel()->resetUsageStats();
234            $this->helper->getRephraseModel()->resetUsageStats();
235            $this->helper->getEmbeddingModel()->resetUsageStats();
236            $result = $this->helper->askChatQuestion($q, $history);
237            $this->colors->ptln("Interpretation: {$result['question']}", Colors::C_LIGHTPURPLE);
238            $history[] = [$result['question'], $result['answer']];
239            $this->printAnswer($result);
240        }
241    }
242
243    /**
244     * Print information about the available models
245     *
246     * @return void
247     */
248    protected function models()
249    {
250        $result = (new ModelFactory($this->conf))->getModels();
251
252        $td = new TableFormatter($this->colors);
253        $cols = [30, 20, 20, '*'];
254        echo "==== Chat Models ====\n\n";
255        echo $td->format(
256            $cols,
257            ['Model', 'Token Limits', 'Price USD/M', 'Description'],
258            [Colors::C_LIGHTBLUE, Colors::C_LIGHTBLUE, Colors::C_LIGHTBLUE, Colors::C_LIGHTBLUE]
259        );
260        foreach ($result['chat'] as $name => $info) {
261            echo $td->format(
262                $cols,
263                [
264                    $name,
265                    sprintf(" In: %7d\nOut: %7d", $info['inputTokens'], $info['outputTokens']),
266                    sprintf(" In: %.2f\nOut: %.2f", $info['inputTokenPrice'], $info['outputTokenPrice']),
267                    $info['description'] . "\n"
268                ],
269                [
270                    $info['instance'] ? Colors::C_LIGHTGREEN : Colors::C_LIGHTRED,
271                ]
272            );
273        }
274
275        $cols = [30, 10, 10, 10, '*'];
276        echo "==== Embedding Models ====\n\n";
277        echo $td->format(
278            $cols,
279            ['Model', 'Token Limits', 'Price USD/M', 'Dimensions', 'Description'],
280            [Colors::C_LIGHTBLUE, Colors::C_LIGHTBLUE, Colors::C_LIGHTBLUE, Colors::C_LIGHTBLUE, Colors::C_LIGHTBLUE]
281        );
282        foreach ($result['embedding'] as $name => $info) {
283            echo $td->format(
284                $cols,
285                [
286                    $name,
287                    sprintf("%7d", $info['inputTokens']),
288                    sprintf("%.2f", $info['inputTokenPrice']),
289                    $info['dimensions'],
290                    $info['description'] . "\n"
291                ],
292                [
293                    $info['instance'] ? Colors::C_LIGHTGREEN : Colors::C_LIGHTRED,
294                ]
295            );
296        }
297
298        $this->colors->ptln('Current prices may differ', Colors::C_RED);
299    }
300
301    /**
302     * Handle a single, standalone question
303     *
304     * @param string $query
305     * @return void
306     * @throws Exception
307     */
308    protected function ask($query)
309    {
310        $result = $this->helper->askQuestion($query);
311        $this->printAnswer($result);
312    }
313
314    /**
315     * Get the pages that are similar to the query
316     *
317     * @param string $query
318     * @return void
319     */
320    protected function similar($query)
321    {
322        $langlimit = $this->helper->getLanguageLimit();
323        if ($langlimit) {
324            $this->info('Limiting results to {lang}', ['lang' => $langlimit]);
325        }
326
327        $sources = $this->helper->getEmbeddings()->getSimilarChunks($query, $langlimit);
328        $this->printSources($sources);
329    }
330
331    /**
332     * Run the maintenance tasks
333     *
334     * @return void
335     */
336    protected function runMaintenance()
337    {
338        $start = time();
339        $this->helper->getStorage()->runMaintenance();
340        $this->notice('Peak memory used: {memory}', ['memory' => filesize_h(memory_get_peak_usage(true))]);
341        $this->notice('Spent time: {time}min', ['time' => round((time() - $start) / 60, 2)]);
342
343        $data = $this->helper->getRunData();
344        $data['maintenance ran at'] = dformat();
345        $this->helper->setRunData($data);
346    }
347
348    /**
349     * Recreate chunks and embeddings for all pages
350     *
351     * @return void
352     */
353    protected function createEmbeddings($clear)
354    {
355        [$skipRE, $matchRE] = $this->getRegexps();
356
357        $data = $this->helper->getRunData();
358        $lastEmbedModel = $data['embed used'] ?? '';
359
360        if(
361            !$clear && $lastEmbedModel &&
362            $lastEmbedModel != (string) $this->helper->getEmbeddingModel()
363        ){
364            $this->warning('Embedding model has changed since last run. Forcing an index rebuild');
365            $clear = true;
366        }
367
368        $start = time();
369        $this->helper->getEmbeddings()->createNewIndex($skipRE, $matchRE, $clear);
370        $this->notice('Peak memory used: {memory}', ['memory' => filesize_h(memory_get_peak_usage(true))]);
371        $this->notice('Spent time: {time}min', ['time' => round((time() - $start) / 60, 2)]);
372
373
374        $data['embed ran at'] = dformat();
375        $data['embed used'] = (string) $this->helper->getEmbeddingModel();
376        $this->helper->setRunData($data);
377    }
378
379    /**
380     * Dump TSV files for debugging
381     *
382     * @return void
383     */
384    protected function tsv($vector, $meta)
385    {
386
387        $storage = $this->helper->getStorage();
388        $storage->dumpTSV($vector, $meta);
389        $this->success('written to ' . $vector . ' and ' . $meta);
390    }
391
392    /**
393     * Print the given detailed answer in a nice way
394     *
395     * @param array $answer
396     * @return void
397     */
398    protected function printAnswer($answer)
399    {
400        $this->colors->ptln($answer['answer'], Colors::C_LIGHTCYAN);
401        echo "\n";
402        $this->printSources($answer['sources']);
403        echo "\n";
404        $this->printUsage();
405    }
406
407    /**
408     * Print the given sources
409     *
410     * @param Chunk[] $sources
411     * @return void
412     */
413    protected function printSources($sources)
414    {
415        foreach ($sources as $source) {
416            /** @var Chunk $source */
417            $this->colors->ptln(
418                "\t" . $source->getPage() . ' ' . $source->getId() . ' (' . $source->getScore() . ')',
419                Colors::C_LIGHTBLUE
420            );
421        }
422    }
423
424    /**
425     * Print the usage statistics for OpenAI
426     *
427     * @return void
428     */
429    protected function printUsage()
430    {
431        $chat = $this->helper->getChatModel()->getUsageStats();
432        $rephrase = $this->helper->getRephraseModel()->getUsageStats();
433        $embed = $this->helper->getEmbeddingModel()->getUsageStats();
434
435        $this->info(
436            'Made {requests} requests in {time}s to models. Used {tokens} tokens for about ${cost}.',
437            [
438                'requests' => $chat['requests'] + $rephrase['requests'] + $embed['requests'],
439                'time' => $chat['time'] + $rephrase['time'] + $embed['time'],
440                'tokens' => $chat['tokens'] + $chat['tokens'] + $embed['tokens'],
441                'cost' => $chat['cost'] + $chat['cost'] + $embed['cost'],
442            ]
443        );
444    }
445
446    /**
447     * Interactively ask for a value from the user
448     *
449     * @param string $prompt
450     * @return string
451     */
452    protected function readLine($prompt)
453    {
454        $value = '';
455
456        while ($value === '') {
457            echo $prompt;
458            echo ': ';
459
460            $fh = fopen('php://stdin', 'r');
461            $value = trim(fgets($fh));
462            fclose($fh);
463        }
464
465        return $value;
466    }
467
468    /**
469     * Read the skip and match regex from the config
470     *
471     * Ensures the regular expressions are valid
472     *
473     * @return string[] [$skipRE, $matchRE]
474     */
475    protected function getRegexps()
476    {
477        $skip = $this->getConf('skipRegex');
478        $skipRE = '';
479        $match = $this->getConf('matchRegex');
480        $matchRE = '';
481
482        if ($skip) {
483            $skipRE = '/' . $skip . '/';
484            if (@preg_match($skipRE, '') === false) {
485                $this->error(preg_last_error_msg());
486                $this->error('Invalid regular expression in $conf[\'skipRegex\']. Ignored.');
487                $skipRE = '';
488            } else {
489                $this->success('Skipping pages matching ' . $skipRE);
490            }
491        }
492
493        if ($match) {
494            $matchRE = '/' . $match . '/';
495            if (@preg_match($matchRE, '') === false) {
496                $this->error(preg_last_error_msg());
497                $this->error('Invalid regular expression in $conf[\'matchRegex\']. Ignored.');
498                $matchRE = '';
499            } else {
500                $this->success('Only indexing pages matching ' . $matchRE);
501            }
502        }
503        return [$skipRE, $matchRE];
504    }
505}
506