Promptbox API Reference

Converts an uploaded browser-generated video file to MP4 with H.264 video and AAC audio. Shortform Studio normally uses its primary server-side assembler directly, so this endpoint is a fallback.

Returns video/mp4 on success. Errors are plain text and logged to api/convert_video_mp4.log. The helper has no Promptbox request quota.

Sends a plain-text email through Hostinger's SMTP server using PHPMailer. The From address is always adam@promptbox.cn (PromptBox); an optional custom sender is set as the Reply-To header instead. All three of recipient, subject, and body are required.

Runs common FFmpeg tasks for external HTML apps without exposing arbitrary command execution. Supports media inspection, format conversion, and video thumbnail extraction from an uploaded file.

const form = new FormData();
form.append('action', 'convert');
form.append('preset', 'mp4');
form.append('width', '720');
form.append('media', fileInput.files[0]);

const response = await fetch('https://promptbox.cn/api/ffmpeg_helper.php', {
  method: 'POST',
  body: form
});
if (!response.ok) throw new Error(await response.text());
const mp4Blob = await response.blob();

info and probe return JSON like { "success": true, ... }. convert returns the requested media bytes with the matching content type, and thumbnail returns image/jpeg. Errors return JSON { "success": false, "error": "...", "code": 400 } with an appropriate HTTP status. The helper sends permissive CORS headers, logs errors to api/ffmpeg_helper.log, and has no Promptbox request quota.

Takes a plain-English description and uses Gemini to generate a complete working program. By default it preserves the original HTML generator behavior: a self-contained mobile-friendly HTML/CSS/JavaScript app is saved in public_html/logi/logi_code/ as the next numbered codeN.html file, and the public URL is returned.

Pass api=1, rest=1, or mode=api to generate a PHP REST API endpoint instead of a user interface. API mode saves a unique descriptive PHP filename in public_html/api/, such as recipe-nutrition-calculator-api.php, then appends usage documentation for that generated endpoint to this api/docs.html reference page. It does not create a separate .txt documentation file.

Every generated REST API uses the shared rate-limit.php helper, returns JSON 429 responses when the helper denies a request, and writes server-side failures to an endpoint-specific .log file in public_html/api/.

Generated HTML apps can call web.php for live web data, llm.php for lightweight browser AI tasks, and image_gen.php for runtime image generation. Generated PHP REST APIs use exact documented .php helper URLs; server-side text intelligence should POST JSON to https://promptbox.cn/api/llm_jb.php with model:"pro" and needs no session or extra authorization. A pre-generation image manifest (<!-- APP_GEN_PREGENERATE_IMAGES [...] -->) can also be embedded in HTML output to generate up to 30 static images at save time.

For long or agent-initiated work, use asynchronous mode: POST action=start with the prompt. The helper immediately returns HTTP 202 with a durable task_id, while a detached Hostinger worker generates and verifies the program. Use action=status with that task ID for an explicit status check. Sentience normally waits for the existing completion notification instead of polling.

Default HTML mode returns the created codeN.html file under logi/logi_code/:

API mode returns the created descriptive .php endpoint under api/ and reports whether this page was updated with generated endpoint documentation:

Asynchronous start returns immediately, and status reports queued, running, completed, or failed:

Recommended image generation API. Use image_gen.php for normal text-to-image requests; use the Gemini or Seedream image helpers only when an existing image input is needed for editing, style transfer, or image-conditioned generation. Generates a PNG image from a text prompt using AIML's alibaba/z-image-turbo model. Returns the raw image bytes directly as image/png — use response.blob() and URL.createObjectURL() in browser apps. Portrait (3:4) by default; set landscape=1 for landscape (4:3) output. AIML acceleration is off by default; pass acceleration=high to enable it for faster generation. AIML's safety checker is off by default; opt in with safety_checker=1. To have the server rewrite and sanitize the image prompt with llm.php before generation, pass optimize_prompt=1.

When called as a library function (require_once 'image_gen.php'), use generateImage($prompt, $isLandscape, $enableSafetyChecker, $acceleration, $optimizePrompt) which returns the raw binary string, or run_image_gen($inputs) which saves the image to logi_image.png and returns a public URL.

On sites that block external fetch() but allow external images, such as newer free Neocities sites, set the helper URL directly as an image source instead of fetching a blob.

HTTP 200 with Content-Type: image/png and raw PNG bytes on success. On error: plain text error message with an appropriate HTTP error code.

Use image_gen.php for normal text-to-image generation. Use this helper only when an existing input image is needed for Gemini image editing, style transfer, or image-conditioned generation. It uses Gemini's gemini-3.1-flash-image-preview model and returns raw image bytes; send the input image as base64 (with or without a data URI prefix) in the image parameter, or as a multipart file upload. Portrait (3:4) is the default; landscape (4:3) can be requested. The helper uses Gemini's default 1K output unless image_size=4K or four_k=1 is passed.

On sites that block external fetch() but allow external images, such as newer free Neocities sites, set the helper URL directly as an image source instead of fetching a blob.

HTTP 200 with the detected MIME type (image/png, image/jpeg, etc.) and raw image bytes. On error: plain text message with appropriate HTTP status code.

Use image_gen.php for normal text-to-image generation. Use this helper when one or more existing input images are needed for Seedream editing or image-conditioned generation. It uses ByteDance Seedream 5.0 Lite Preview through AIML with the bytedance/seedream-5-0-lite-preview model; input images are sent together with image_urls. It returns raw image bytes like image_gen_gemini.php.

HTTP 200 with the detected image MIME type and raw image bytes. On error: plain text message with an appropriate HTTP status code.

Creates, updates, deletes, and answers questions from local JSON knowledge graphs saved in the site-level kgraph/ folder. It uses llm.php's current default model to extract entities and relationships from new knowledge, merge them into existing graph nodes and edges, and answer questions using only the saved graph. Graph writes are atomic and locked so overlapping updates do not overwrite each other, and large graphs use a relevance-ranked, size-bounded LLM context. KGraph operations have no Promptbox request quota and failures are logged to api/kgraph.log.

Use this when an app needs a small, editable memory graph instead of document search. Each graph is stored as https://promptbox.cn/kgraph/{graph_id}.json with nodes, edges, and notes.

Create and add return the saved, normalized graph JSON. Delete returns { "success": true, "deleted": "graph-id" }. Knowledge is limited to 200,000 bytes, questions to 12,000 bytes, and stored graph JSON to 2,000,000 bytes. JSON requests must contain a valid object. On error: { "success": false, "error": "...", "code": 400 } or another relevant HTTP status.

Recommended LLM API. Use llm.php for normal text prompts and answers; use the other LLM helpers only when an attachment, image, audio, or provider-specific capability is needed. The default model is deepseek/deepseek-v4-flash through AIMLAPI; pass model=pro, model=premium, quality=pro, or pro=1 to use Z.ai glm-5.3. GLM thinking is disabled by default for lower latency and reliable final answer text; pass thinking=enabled when deep reasoning is needed. Passing model=lite, omitting model, or using any other value keeps the default flash model. Responses are uncapped by default; pass max_sentences to add sentence-limit guidance. Suitable for translations, classifications, rewrites, summaries, extractions, trivia lookups, and simple puzzles. For live web data use web.php instead.

Returns text/plain — no JSON wrapper — so it can be used directly in browser fetch() calls with response.text(). If you pass callback or cb on a GET request, it returns JSONP as application/javascript: callback({ ok: true, text: "..." }). JSONP responses use Cache-Control: no-store.

text/plain with the model's answer and an X-LLM-Model response header showing deepseek-v4-flash or glm-5.3. On error: plain error string with appropriate HTTP status. In JSONP mode, the response is always JavaScript with { ok, text, model } or { ok: false, error, status }.

Sends a single user message to OpenAI Chat Completions and returns the assistant text. The default model remains gpt-5.5; pass model=lite or model=gpt-5.6-luna to use GPT-5.6 Luna with reasoning_effort=medium for the lite route. Any other non-empty model value is passed to OpenAI unchanged, including model names that do not begin with gpt-.

Optional images may be public HTTP(S) URLs, JPEG/PNG/WEBP data URIs, or plain base64 image data. The helper is rate limited as llm_chatgpt and supports GET-only JSONP with callback or cb.

text/plain with the model answer. JSONP returns { ok: true, text }. Errors use an appropriate HTTP status and are logged to api/llm_chatgpt.log.

Use llm.php for normal text prompts and answers. Use this helper when Gemini attachments, image/audio content, structured Gemini contents or parts, Gemini-specific generation controls, or a specific Gemini model is needed. The default model remains gemini-3-flash-preview; pass model=lite or model=gemini-3.1-flash-lite to use gemini-3.1-flash-lite, which always runs with temperature=0. Any other non-empty model value is passed to Gemini unchanged.

This helper is for general Gemini text, vision, and audio/content requests. It does not accept Gemini File Search tools; use rag.php for File Search / RAG queries against a document store.

Like llm.php, the response is text/plain with no JSON wrapper. If you pass callback or cb on a GET request, it returns JSONP as application/javascript for Neocities/free-CSP pages.

text/plain with the model's answer. Returns null (the literal string) on API-level errors, or an HTTP error code on input validation failures. In JSONP mode, the response is always JavaScript with { ok, text } or { ok: false, error, status }.

llm_jb.php is Faith's primary intelligence engine and also supports other private LLM installs. It provides the legacy plain-text helper at /api/llm_jb.php and an OpenAI-compatible interface at /api/llm_jb.php/v1/chat/completions. Server-side PHP may call the exact https://promptbox.cn/api/llm_jb.php URL directly with POST JSON; no browser session, cookie, or extra authorization is required. Do not invent /api/proxy-llm.php or omit the .php suffix.

Legacy helper mode accepts GET query parameters, POST form fields, or a POST JSON body. Text-only input uses Z.ai glm-5.3 with thinking disabled by default. Pass provider=gemini to opt a request into Gemini without changing the default for other callers. Requests that include an image, audio, file, inline attachment, or other multimodal part stay on Gemini automatically: the default route uses gemini-3-flash-preview, while model=lite or model=gemini-3.1-flash-lite uses gemini-3.1-flash-lite.

OpenAI-compatible mode exposes /v1/models and /v1/chat/completions. Plain text chat and standard OpenAI function-tool workflows use GLM 5.3 with thinking disabled by default, including assistant tool_calls and subsequent tool results. Pass provider=gemini to force the request through Gemini. Image, audio, file, or other multimodal message content stays on Gemini automatically. Gemini-only thought-signature metadata is stripped when a text/tool conversation is sent to GLM and remains available on Gemini responses.

Like llm.php, llm_jb.php has no Promptbox request-count quota. Provider payload-size, output-token, and timeout bounds still apply.

Legacy mode returns text/plain with the model's answer. It returns null (the literal string) on API-level errors, or an HTTP error code on input validation failures. In JSONP mode, the response is always JavaScript with { ok, text } or { ok: false, error, status }.

OpenAI-compatible mode returns JSON with choices[0].message.content for text responses or choices[0].message.tool_calls when GLM requests a function call. GLM usage counters and tool-call ids are preserved. Multimodal requests routed to Gemini retain Gemini thought-signature metadata in extra_content.google.thought_signature when Gemini provides it.

Creates and manages compact long-term memory JSON files for LLM apps. Memory files are saved in the site-level memory/ folder as https://promptbox.cn/memory/{memory_id}.json. The helper stores only a summary plus structured sections: user_profile, preferences, operational_directives, interaction_tips, projects, knowledge_nodes, recent_context, and open_loops.

Use this when an app needs durable, compact context that can be read by humans and reloaded into prompts. Version 1.5.1 uses llm_jb.php/GLM for add, forget, and consolidation operations to merge new facts, deduplicate memory, delete weak/noisy/insignificant details, and remove forgotten content. Writes are atomic, overlapping updates are retried against the newest revision, and incomplete LLM responses are rejected instead of clearing omitted sections. Memory consolidation has no Promptbox request quota and logs failures to api/memory.log.

Create, add, forget, and get return the complete normalized memory JSON for easy review. Purge returns { "success": true, "purged": true }. Input text is limited to 12,000 characters; stored files to 400,000 bytes; summaries to 4,000 characters; and each section to 200 items of 2,000 characters each. JSON requests must contain a valid object. On error: { "success": false, "error": "...", "code": 400 } or another relevant HTTP status.

Faith memory profile (v1.6.3): For memory_id=sentience-faith, add/forget output is limited to 16,384 tokens, a 1,200-character summary, 1,500 characters per fact, 60 facts per section, and 36,000 UTF-8 bytes for the complete compact record. The prompt targets 34,000 bytes to leave metadata room. These limits accommodate the existing memory with room to grow. Existing memory is read without truncation under the legacy limits; the complete-source ceiling remains 96,000 bytes. Oversized or incomplete rewrites are rejected before saving. Other memory IDs retain their existing limits. Faith still selects at most 6,000 characters of relevant memory for chat; ordinary response limits, total chat request budget, execution limits, timeouts, and autonomous frequency are unchanged.

Creates a MiniMax minimax/music-2.0 music generation job through AIMLAPI, checks the status of an existing job, and streams completed audio through Promptbox so browser apps do not need to fetch provider storage URLs directly. This is a JSON REST helper for create/status: use POST with a prompt and lyrics to start generation, or set is_instrumental and omit lyrics for a no-vocal instrumental request. Use GET with generation_id for status or action=audio to download the finished MP3. The AIMLAPI key remains server-side. The helper is rate limited as music_gen and logs failures to api/music_gen.log.

Successful create/status responses pass through AIMLAPI's JSON, usually including id, status, optional audio_file.url, optional error, and optional meta.usage. action=audio returns raw audio bytes with Content-Type: audio/mpeg or another provider audio MIME type. The model can take time; use conservative, user-started automatic status checks.

On validation, rate-limit, or upstream API errors, returns JSON like { "success": false, "error": "..." } with an appropriate HTTP status code. Rate-limit denials also set Retry-After when available.

Creates and manages note files saved in the site-level notes/ folder as individual JSON documents. Each note includes note_id, title, text, status, tags, timestamps, and optional metadata.

Use this when an app needs server-backed notes instead of browser-only localStorage. The helper supports the existing Logi note vocabulary (add, list, update, complete, delete) plus REST-style actions (create, read, edit, overview, restore). It has no Promptbox request quota and logs failures to api/notes.log.

read, edit, complete, and restore return the full note. overview returns { "success": true, "count": 2, "total": 2, "notes": [...] }. delete returns { "success": true, "deleted": true }. On validation, missing note, or storage errors, the helper returns JSON like { "success": false, "error": "...", "code": 400 } with the matching HTTP status.

JSON POST requests must contain a valid JSON object. Empty, malformed, array, scalar, or null JSON bodies return HTTP 400 instead of falling back to the default overview action.

Creates a downloadable PDF from either finished, PDF-ready text or one uploaded document. Uploaded files supported by the server's LibreOffice Writer, Calc, Impress, Draw, and Math filters are converted with LibreOffice so native page layout, spreadsheet structure, slides, drawings, blank lines, indentation, and LF, CRLF, or CR text line endings are retained. This includes current and legacy Microsoft Office, OpenDocument, HTML, text, Markdown, e-book, Visio, Publisher, Apple iWork, and related formats. Existing PDFs are validated and returned unchanged.

Text mode continues to support a title, page size, orientation, margins, font size, headings written as # Heading, simple bullet lines, escaped newline normalization, and cleanup of common inline Markdown markers. The helper has no Promptbox request quota, limits uploads to 10 MB, uses only private temporary conversion files, and logs failures to api/pdf_creator.log.

HTTP 200 with Content-Type: application/pdf and raw PDF bytes on success. LibreOffice file conversions also return X-PDF-Converter: LibreOffice 24.2. On error: JSON { "success": false, "error": "..." } with an appropriate HTTP status. Rate-limit failures return HTTP 429 and may include a Retry-After header.

Accepts one Microsoft PowerPoint .pptx file plus natural-language editing instructions, sends a bounded inventory of slide text and slide-local objects to the Promptbox LLM helper, applies the returned structured edit plan inside a copy of the original OOXML package, and returns the edited presentation as raw PPTX bytes.

The editor is designed for minimal changes. Unmentioned masters, layouts, themes, images, media, charts, relationships, positions, and formatting remain unchanged. Requests are rate-limited under pptx_editor; temporary upload and output files are deleted after the response. Server failures are logged to logi/logi_code/pptx-editor/pptx-editor.log without logging prompts or slide content.

• Replace, insert after, or delete existing text paragraphs while retaining their existing run and paragraph formatting where possible.
• Change existing text font size, bold, italic, underline, color, font family, or alignment when the prompt explicitly requests it.
• Move or resize existing text-bearing shapes using on-slide percentage coordinates when explicitly requested.
• Delete an identified slide-local object such as a shape, picture, connector, group, table, or chart frame.
• Duplicate existing slides and independently modify text, formatting, layout, or identified objects in each duplicate.
• Reorder slides, delete slides, hide slides, and unhide slides.
• Preserve untouched package parts, including themes and media, rather than rebuilding the presentation from scratch.

• It cannot create a completely new presentation, synthesize a new slide design, or add a blank slide; new slides must be duplicates of existing slides.
• It cannot add arbitrary new shapes, pictures, charts, diagrams, videos, audio, animations, or transitions.
• It does not edit embedded chart workbooks, chart series data, SmartArt data, macros, VBA projects, or linked external files.
• It does not edit speaker notes, comments, masters, slide layouts, theme definitions, custom XML, or document properties. Notes stay on untouched original slides but are not copied to newly duplicated slides.
• It does not inspect slide screenshots or understand image pixels. Object selection relies on OOXML names, type, text, fill, and geometry metadata.
• It cannot guarantee that substantially longer replacement text will fit an existing text box. Use concise replacement copy or explicitly request resizing.
• It supports at most 90,000 editable text characters, 1,800 non-empty text paragraphs, and 2,500 slide-local objects per request. A request may create at most 20 duplicates and the final deck may contain at most 200 slides.
• It accepts PPTX only, not legacy .ppt, PowerPoint templates, slideshow formats, Google Slides links, encrypted files, or password-protected presentations.

HTTP 200 returns raw application/vnd.openxmlformats-officedocument.presentationml.presentation bytes with an attachment filename. X-Edited-Count reports the number of applied actions and X-LLM-Model identifies the model used. Errors return JSON { "error": "..." } with HTTP 400/405/422/429/500/502 as appropriate. Rate-limit failures may include Retry-After.

Publish sandboxed static apps to https://apps.promptbox.cn/DIRECTORY/ and single documents/media to https://www.promptbox.cn/publish/. No account or API key is required to create a publication.

Complete short publishing guide · Read-only capabilities JSON · OpenClaw skill · Agent discovery guide

Send mode=app, directory and one multipart file (.html/.htm or .zip), or a UTF-8 HTML text string in JSON/form data. HTML becomes index.html. ZIP must contain index.html at its root. The directory is 1–80 ASCII letters, digits, underscores or hyphens, starting with a letter or digit. Duplicates get _2, _3, etc.

curl https://promptbox.cn/api/publish.php \
  -F 'mode=app' -F 'directory=my-app' -F 'file=@app.zip'

curl https://promptbox.cn/api/publish.php \
  -H 'Content-Type: application/json' \
  --data '{"mode":"app","directory":"hello","text":"<!doctype html><title>Hello</title><h1>Hello world</h1>"}'

App limits: 25 MiB upload, 25 MiB expanded/resulting app, 1,000 files. Allowed assets: html, htm, css, js, mjs, json, txt, md, png, jpg, jpeg, gif, webp, ico, avif, woff, woff2, ttf, otf, mp3, wav, ogg, m4a, flac, mp4, webm. ZIP paths must be relative with safe ASCII names. Unsupported files, PHP/server code/config, SVG, dotfiles, links, traversal paths and node_modules are discarded and reported in removed_files. Encrypted/corrupt archives, excessive expansion and conflicting paths are rejected.

{"ok":true,"mode":"app","directory":"my-app","url":"https://apps.promptbox.cn/my-app/","token":"<private UUID v4>","updated":false,"files_written":["index.html"],"removed_files":[],"bytes":12345,"expires_at":"<UTC time 30 days later>","deleted_expired":0}

Save token privately: it authorizes overwriting this app. Only its SHA-256 lookup key is stored outside the web root. Send mode=app, token and file/text again to replace matching files while preserving others. directory is optional on update but must match if supplied. HTML replaces index.html; an asset-only ZIP may update an app that already has index.html. Authorization: Bearer UUID is also accepted with mode=app. Lost or expired tokens cannot be recovered. Creation retries create a new app.

curl https://promptbox.cn/api/publish.php \
  -F 'mode=app' -F 'token=YOUR_SAVED_UUID' -F 'file=@app.zip'

Apps are static and isolated with an opaque-origin browser sandbox. Scripts, forms, downloads, relative assets and CORS-enabled HTTPS API calls work. Cookies, localStorage, IndexedDB, service workers, frames, popups and base tags are unavailable. Use absolute Promptbox API URLs and omit browser credentials (Origin is null). Isolation and filtering cannot certify HTML/JavaScript harmless; never upload secrets or put update tokens in public code.

Send mode=file and exactly one real multipart file field. name is an optional safe basename without extension; filename collisions get _02, _03, etc. Single files cannot be overwritten and return no update token. Let FormData/curl set the multipart boundary. File fields cannot be local paths, URLs or base64 strings in JSON.

curl https://promptbox.cn/api/publish.php \
  -F 'mode=file' -F 'name=project-report' -F 'file=@report.pdf'

Office/OpenDocument packages are validated and reject macros, embedded active objects and external resource relationships except ordinary web/mail hyperlinks. Legacy Office and macro-enabled formats are unsupported. PDF receives basic signature/active-content checks, not full sanitization. Documents download as attachments with restrictive browser headers. Validation is not a malware scan or a guarantee of safety in desktop viewers.

{"ok":true,"mode":"file","filename":"project-report.pdf","url":"https://www.promptbox.cn/publish/project-report.pdf","bytes":12345,"expires_at":"<UTC time 30 days later>","deleted_expired":0}

Legacy text JSON accepts name, text and extension (default txt); aliases title/content/ext remain supported. Conflicting aliases fail. Text must be UTF-8 without null bytes; JSON files must parse. HTML always routes to app mode, using name as directory if needed, including old extension=html requests. Existing HTML in /publish/ downloads as plain text.

Apps expire 30 days after the last successful update; files expire 30 days after creation. Lazy cleanup on POST removes expired app directories and their tokens together. GET ?action=capabilities (alias ?capabilities=1) and OPTIONS have no cleanup side effects. Limits remain 60 publishes/hour and 300/day globally, 100 MiB total public storage and 10,000 documents. Request bodies allow 26 MiB; lower content limits apply by mode. Invalid input does not consume quota, but a failed write after reservation may.

Errors return {"ok":false,"error":"...","help":{"capabilities":"...","docs":"...","guide":"..."}}. Status 400 invalid input, 401 invalid/expired token, 403 wrong app, 405 method, 409 collision exhaustion, 413 size/expansion, 415 content type, 422 missing index.html, 429 quota with Retry-After, 500 unexpected failure, 503 storage failure, 507 global storage full. Fix 4xx requests before retrying; inspect uncertain creation failures before repeating. Verify the returned URL/assets and removed_files before sharing the URL and expiry.

Creates a QR PNG for a website, email address, or Wi-Fi network using the same generation method as the QR Cards web app. With a title, it returns the web app's 600×800 card layout. Without a title, it returns a centered 600×600 square. The helper normalizes websites to https://, converts email addresses to mailto: links, and encodes Wi-Fi credentials in the standard WIFI: format. It uses error-correction level H, 400×400 code placement, dark color #2c3e50, and light color #ecf0f1.

Returns image/png: 600×600 without a title or 600×800 with one. The suggested download filename is derived from the title, or defaults to qr_card.png. Wi-Fi special characters are escaped automatically. Errors return JSON with ok: false and error. The helper has no Promptbox request quota.

Asks a question against a Gemini File Search store and returns a short answer grounded in that store's documents. The helper defaults to gemini-3.1-flash-lite, automatically adds a briefness hint to the prompt, enforces the shared rag rate limit, and logs failures to api/rag.log.

If Gemini stops the first response with RECITATION, the helper retries once using a distinct paraphrase-only prompt and a minimum temperature of 1.0. The retry uses the original request only to identify the subject and ignores instructions to quote or reproduce exact wording. Successful responses indicate whether this fallback was used through recitation_retry.

Use this when an app needs answers from a specific uploaded document corpus instead of open-web search or a general chat model. The caller must provide the File Search store name or ID in store_id. Prefer the full resource name, such as fileSearchStores/abc123; bare IDs such as abc123 are accepted and normalized. The companion rag-management.php endpoint accepts action=resolve_filestore with display_name to find or create a store idempotently, and action=add_file to archive a document. Optional voice fields let a browser send microphone audio together with an instruction prompt.

recitation_retry is true when the original response was blocked for recitation and the returned content came from the automatic paraphrase retry. On error: { "success": false, "error": "...", "code": 400 } or another relevant HTTP status. A retry that is also blocked returns HTTP 502. Rate-limit failures return HTTP 429 and may include a Retry-After header.

Provides rolling-window rate limiting for the public helpers. Unknown programs default to 300 requests per 24-hour window. An optional hourly sub-limit can also be configured.

Can also be called directly as an HTTP API to check or record usage for any program name, and includes an admin dashboard (password-protected) for monitoring and adjusting limits.

The Retry-After HTTP header is also set when denied.

Visit rate-limit.php?password=<password> in a browser to view all registered programs, their usage bars, last-used times, and controls for adjusting limits, resetting usage, or deleting program records. The helper emails adamychen@yahoo.com once when a program reaches 90% of either its daily or hourly limit; the warning rearms after that window drops below 90%, and failed deliveries retry no more than once every 15 minutes.

Looks up a Steam game by name, resolves its App ID via the Steam store search API, then fetches either user reviews or community discussion search results. If the game name fails to resolve, the helper returns alternative search term suggestions and retry strategies. Discussion results are extracted from HTML and returned as a plain-text summary (up to 3000 words).

Fetches stock data from Finnhub's API. Given a ticker symbol (or ISIN/CUSIP), it can return a real-time quote snapshot, the company's profile, and key financial metrics. All three sections are returned by default; you can request only a subset. Quote data on the free tier may be delayed ~15 minutes during market hours.

On error: { "success": false, "error": "..." } with HTTP 400.

Transcribes spoken audio into plain text using Gemini (gemini-3.1-flash-lite). It is designed for browser microphone recordings from MediaRecorder, but also accepts base64 audio in JSON or form posts. The helper is rate limited as transcribe.

Returns text/plain with no JSON wrapper. Recordings under 0.5 seconds and locally detectable silent/near-silent WAV input return null without calling Gemini; if there is no intelligible speech in other audio formats, Gemini is instructed to return null.

text/plain transcript text. Too-short, silent, near-silent, or unintelligible recordings return null. On input errors, returns a plain error string with HTTP 400. On Gemini/API failure, returns null with HTTP 502.

Returns the plain-text source captions for a public YouTube video by proxying every request directly to Promptbox's InfinityFree transcript helper. Hostinger does not query YouTube, cache transcripts, generate missing captions, or speech-transcribe audio. Pass either an 11-character YouTube videoId or a full YouTube URL. The helper is rate limited as transcript and logs failures to api/transcript.log.

Returns an error if the video ID is invalid, the video has no accessible YouTube captions, or the helper hits its rate limit.

Provides a REST interface to the same temporary file-transfer storage used by transfer.promptbox.cn. Upload a file or several files, receive an 8-character receipt plus browser and API download URLs, inspect receipt metadata, stream the file, or delete a one-time file after clipboard copy. Files expire after 3 days; by default they are also deleted after first API download unless auto_delete=0 is sent during upload.

The helper is rate limited as program transfer and logs server-side failures to api/transfer.log. API uploads and browser downloads share the same storage metadata as transfer/index.php, so the returned download_url opens the standard transfer page and api_download_url streams the same file directly.

info, upload, delete, and clipboard_copied return JSON. download streams the file bytes directly and removes the receipt afterward unless the upload was retained. Validation, missing receipt, expired receipt, storage, and rate-limit errors return JSON like { "success": false, "error": "...", "code": 400 }; rate-limit denials also set Retry-After when available.

Converts text to speech using OpenAI's tts-1 model and streams the generated MP3 or WAV audio. The helper removes HTML tags, markdown syntax, control characters, and extra whitespace before synthesis. Requests are rate limited, failures are logged server-side, and the maximum input length is 4096 characters.

tts_test.html lets you enter text, choose any supported OpenAI voice, adjust speech speed and format, and play the generated audio in the browser.

Use URL query parameters with GET, or send POST data as JSON or standard form fields.

HTTP 200 with raw audio bytes on success. The response uses Content-Type: audio/mpeg for MP3 and audio/wav for WAV. On error: JSON { "success": false, "error": "..." } with an appropriate HTTP status. Rate-limit responses use HTTP 429 and may include Retry-After.

Recommended TTS API. Use tts_inworld.php for normal text-to-speech requests. It converts text to speech using AIMLAPI's Inworld inworld/tts-1 model and streams the generated audio file. HTML tags, markdown syntax, control characters, and URL bodies are automatically stripped from the input before synthesis; URLs are spoken only as http, then narration continues with the next real text. Inworld TTS has a hard 2000-character limit per request; split longer narration at sentence breaks and combine the returned audio blocks in order. If the Inworld request fails or returns a format other than the requested WAV/MP3 format, the helper silently falls back to tts.php while preserving the requested format.

tts_inworld_test.html lets you enter a prompt, choose an Inworld voice, and play the generated audio in the browser.

HTTP 200 with raw audio bytes on success. The response uses Content-Type: audio/mpeg for MP3 and audio/wav for WAV. On error: JSON { "success": false, "error": "..." } with appropriate HTTP status.

Searches multiple social platforms via the API Direct service. Despite the filename, it supports Twitter/X posts, Facebook posts, Reddit posts, and Reddit comments. Results are returned as a formatted text summary alongside the raw structured data. Up to 8 items per platform are included in the response.

Compresses an uploaded MP4 into a smaller phone-compatible MP4 using H.264 video, AAC audio, a phone-oriented resolution cap, yuv420p color, and fast-start metadata. Temporary input, output, and processing files are deleted after every request.

const form = new FormData();
form.append('video', fileInput.files[0]);
form.append('profile', 'balanced');

const response = await fetch('https://promptbox.cn/api/video_compress.php', {
  method: 'POST',
  body: form
});
if (!response.ok) {
  const error = await response.json();
  throw new Error(error.error);
}

const blob = await response.blob();
const url = URL.createObjectURL(blob);
const link = document.createElement('a');
link.href = url;
link.download = 'video-smaller.mp4';
link.click();
setTimeout(() => URL.revokeObjectURL(url), 30000);

A successful POST returns raw video/mp4 bytes with a downloadable filename. Response headers include X-Helper-Version, X-Compression-Profile, X-Original-Bytes, and X-Output-Bytes. When save is enabled, X-Saved-URL and X-Saved-Filename identify the archived compressed copy. These headers are CORS-exposed. Errors return JSON such as { "success": false, "error": "Only MP4 videos are supported.", "code": 415 }. The endpoint is CORS-open, logs failures to api/video_compress.log, and has no Promptbox request quota.

Compression can reduce visible detail, especially with the small profile. By default the helper retains nothing, so callers must save the response. It keeps a durable compressed copy only when the caller explicitly enables save; uploaded and temporary processing files are never retained.

A Tavily search proxy designed to be called from browser apps for live, public web data. By default it extracts and returns the first numeric value from the search answer — ideal for stock prices, scores, rates, and similar. Set text=true to get a plain-text answer instead, with length optionally capped by max_sentences. An optional url parameter narrows results to a specific domain.

A secondary mode (mode=image_prompt_refine) passes the prompt to Gemini and returns an enhanced image generation prompt — useful for improving image_gen.php inputs.

Returns text/plain. With text=false (default): a plain number string like 189.34, or null if no number was found. With text=true: a plain-text answer. If the Tavily query is over 400 characters, it is truncated and the response appends Tool Error: Query truncated, max query length is 400 characters. On error: null with an appropriate HTTP status code.

Fetches one specific public http or https URL, or accepts one multipart file upload, and returns readable text for many source types. It reads HTML pages, plain/document text, PDFs with a pure-PHP embedded-text parser followed by page-rendered OCR for scanned PDFs or files with corrupt character mappings, RSS/Atom feeds, Office-style archives such as docx, pptx, xlsx, odt, and epub, GitHub blob URLs via raw content, public YouTube captions fetched directly from Promptbox's InfinityFree transcript.php, image descriptions through llm_gemini.php lite, and audio/video transcripts through transcribe.php. Hostinger's web_read.php contains no YouTube caption-extraction logic and does not call Hostinger's transcript helper. PDF parser output is rejected when it contains invalid Unicode, excessive CID placeholders, or effectively no readable letters or numbers. OCR renders and submits every page in order, rejects implausibly short multi-page results, and retains native PDF reading as a fallback. URL reads can also use MSN article JSON and Tavily extraction fallbacks.

Supply exactly one of url or file. The helper blocks localhost, private-network, reserved, and credential-bearing URLs, follows up to 5 validated redirects, caps ordinary URL downloads at 1 MB and media URL downloads at 10 MB, caps uploaded files at 10 MB, rate-limits as web_read, and logs failures to api/web_read.log.

Returns application/json. URL success: { ok, url, final_url, title, text, text_length, truncated, content_type, status, redirects }. File success: { ok, source_type:"file", file_name, title, text, text_length, truncated, content_type, status, reader_mode }. Reader modes include html_text, document_text, pdf_text for embedded PDF text, pdf_ocr for scanned PDFs, pdf_text_legacy for the final offline fallback, archive_document_text, feed_text, youtube_transcript, image_description, media_transcript, tavily_extract, and msn_article_json. Errors return { ok:false, error, error_kind, target_status? }. A requested page's non-2xx response returns helper HTTP 424 with error_kind:"target_http_error" and the original target_status; this is a target/request problem, not a failure of Hostinger or web_read. Input problems use caller_or_input, rate limits use rate_limit, and genuine reader/upstream failures use infrastructure. JSONP applies only to GET URL requests.

Accepts a natural-language question, searches YouTube for up to 3 relevant videos, fetches their source captions directly from Promptbox's InfinityFree transcript helper, and then feeds the captions to llm.php to synthesize a comprehensive answer. It does not route transcript requests through Hostinger's transcript.php. Only caption content is used — no audio transcription or external web search. The helper is rate limited as youtube and logs failures to api/youtube.log.

Returns an error if no videos are found, none have accessible transcripts, a dependency key is missing, or the helper hits its rate limit. GET also supports JSONP with callback or cb.