Changelog
v1.4.2
Released: 2026-08-08
- AI Daily now opens with one big, obvious search box, the same one you use on the homepage: you arrive, you type, done — no panel of options to get past first.
- The filters by company, edition, category, and date are all still there and work exactly as before — they just tuck into a small control now instead of taking over the top of the page.
- The article list was redesigned as a single, wider, airier column: it reads straight down like a blog instead of as a tight grid.
- For now the index simply shows the most recent stories first. The “featured” story comes back once the archive is big enough for singling one out to actually mean something.
v1.4.1
Released: 2026-08-08
- You can now search the AI Daily archive: type and it looks across the headlines, decks, and tags of every published story.
- Filters by company, edition type, category, and date range, plus newest-first or oldest-first sorting. Every combination has its own web address, so you can share it or bookmark it — and it works even with JavaScript off.
- Two new editions alongside the daily story: a Saturday “week in review” and a Sunday “what to watch next week”. Each one states the period it covers, and a Saturday can carry both the recap and a big story that broke that day.
- When a story covers an earlier event, it says so plainly: you see the event date, the publication date, and the stated reason it ran late. It can only reach 14 days back, and the automated routine still publishes same-day news only.
- AI Daily moved up in the menu, right after Find the Best AI, and the homepage now introduces the section.
v1.4.0
Released: 2026-08-08
- New AI Daily section (/blog): one story a day about what happened in AI, in Spanish and English, with its own index and an RSS feed per language.
- Every article shows where it came from: linked sources marked primary or secondary, a “why this matters” block, key takeaways, and estimated reading time.
- Same-day rule: a story only goes out if the event happened that same day (Buenos Aires time). If nothing verifiable happened, nothing is published — an empty day beats a warmed-over story.
- Visible corrections: if an article is corrected after publication, the correction is recorded on the article with its date. Nothing is edited silently.
- Secure automated publishing: the routine that assembles the daily story publishes through a signed, least-privilege channel, cannot publish the same story twice, and logs every attempt.
- Version history back in sync: this page had fallen one release behind, so the missing v1.3.2 entry was restored.
v1.3.2
Released: 2026-08-06
- New guide: Prompts for Kimi — how to use its large context window for long-document Q&A, multi-source research synthesis, and precise document anchoring.
- New guide: AI prompts for data analysis — templates for SQL generation, trend identification, and data interpretation, with explicit schema to avoid hallucinated column names.
- New guide: AI prompts for learning — how to use AI as a tutor calibrated to your level, the Feynman test to check comprehension, and building study plans.
- Sitemap fix: four public routes (/best-ai, /prompts/code, /prompts/image, and /prompts/marketing) were missing from sitemap.xml and are now included.
v1.3.1
Released: 2026-08-04
- Version history caught up: this page now documents v1.2.1, v1.2.2, v1.2.3, and v1.3.0, which were missing since July.
- The voice dictation button now has a short explainer next to it describing what it does, in both modes.
- The “New? See how it works →” button now also appears on Find the Best AI, with its own step-by-step explanation of that mode (detected signals, deterministic scoring, alternatives, and the handoff to the optimizer).
- Copy touch-ups in the analyzer help modal: the steps now mention voice dictation and the step count reads correctly.
v1.3.0
Released: 2026-08-04
- Promptea now has two route-addressable modes: Improve Prompt (the analyzer you know, evolved) and Find the Best AI, a new experience that tells you which AI fits the prompt you already have.
- Find the Best AI runs a 100% deterministic matcher: it detects concrete signals in your prompt (repo paths, source requests, strict JSON, images, translation…), scores each AI against capabilities verified from official docs, distinguishes environments like Claude Code from plain chat, and shows a score, confidence, honest ties, up to 3 alternatives with trade-offs, and advice that cites what it found in your text.
- One click on “Improve my prompt for this AI” transfers your full prompt to the optimizer with the recommended AI, model, and purpose preselected — no copy-paste.
- The optimized prompt no longer starts with the PROMPTEA header: version, model, and purpose now live in result metadata. Each task type gets its own shape: short messages stay natural, repo tasks get objective and validation sections, data requests get schema and rules — and an already-good prompt comes back nearly untouched.
- Voice dictation in both modes: record (up to 2 minutes), we transcribe it with Whisper via Groq, and you ALWAYS review and edit the text before using it. Audio is never stored.
- New themes: Aqua (light, liquid-glass style with Apple accents) and Metro (dark, faithful to Windows 8 Metro: flat tiles, azure #0078D7, Segoe type). The “Old version” theme keeps the previous look for anyone who prefers it. System preference maps to Aqua/Metro and legacy saved themes migrate automatically.
- General feedback is now sent from a modal and stored safely (without your prompt, email, or raw identifiers) instead of opening your email client. Analysis helpfulness feedback keeps working as before.
- Updated privacy page (voice and feedback), typed operational-only telemetry, and the test suite grew from 159 to 249 cases across 22 files.
v1.2.3
Released: 2026-07-30
- New guide: Zero-shot prompting — when instruction-only prompts work and when to switch to few-shot, how to fix inconsistent results, and the constraints that make them reliable.
- New guide: Prompts for GPT — labeled format templates, constraint block patterns, instruction placement for long documents, and self-verification requests.
- New guide: AI brainstorming prompts — techniques to escape generic ideas: banning the obvious, forced diversity angles, cross-domain combinations, and separating generation from evaluation.
- Localization fix: the “Templates” section heading on guide pages now correctly reads “Plantillas” in Spanish.
v1.2.2
Released: 2026-07-23
- New guide: Prompt chaining — how to break complex tasks into sequential AI steps, the three chaining patterns (sequential, branching, looping), and their most common failure modes.
- New guide: AI writing prompts — the five elements a writing prompt needs beyond the topic (audience, tone, format, length, avoidance list) and voice-control techniques.
- New guide: Multimodal prompts — how to write prompts when the input includes images, screenshots, or PDFs, with spatial anchors and output format constraints.
- Guide detail pages now emit page-specific Open Graph and Twitter Card metadata for better share previews.
v1.2.1
Released: 2026-07-16
- New guide: System prompts — how to write persistent instructions for ChatGPT, Claude, and Gemini that shape every response without repeating yourself: persona, scope, format defaults, and escalation rules.
- New guide: AI translation prompts — templates that preserve meaning, tone, and register, with domain glossaries, format preservation, and length drift prevention.
- New guide: Image generation prompts — subject, style, composition, lighting, palette, and negative constraints for consistent results in DALL-E, Imagen, and similar models.
- Guide pages now emit BreadcrumbList JSON-LD alongside the existing FAQPage block, enabling breadcrumb paths in Google search results.
v1.2.0
Released: 2026-07-12
- Full redesign: new typography, four themes (Day, Dusk, Night, and Paper) built on semantic tokens with AA contrast, a calm ambient background, and monospace prompt blocks.
- New adaptive refinement engine: routes each prompt to a strategy (short message, coding, strict JSON, agent task, etc.), protects literals like URLs and file paths, validates output through a quality gate, and always falls back to the deterministic engine on failure.
- Simple prompts no longer receive the full section scaffold: the rewrite adapts to the request's real complexity.
- Improved/Original comparison on the result, with a summary of what changed, assumptions, and open questions when the AI refiner ran.
- Model catalog verified against official docs (2026-07-12): GPT-5.6, Claude Fable 5 / Opus 4.8 / Sonnet 5 / Haiku 4.5, Gemini 3.5 Flash and 3.1 Pro, Grok 4.5, DeepSeek V4, Kimi K2.6, and the Sonar family. Old aliases are now marked legacy.
- Scoring fix: well-formed prompts were being under-rated by detection bugs; the 14 failing calibration cases now pass.
- Accessibility: full keyboard navigation, screen-reader announcements, and prefers-reduced-motion support.
v1.1.6
Released: 2026-07-09
- New guide: Prompts for DeepSeek — code with strict constraints, JSON extraction, and step-by-step analytical reasoning.
- New guide: Few-shot prompting — how to show examples instead of just explaining, with text classification and style rewriting templates.
- New guide: Prompt quality scoring — five-dimension self-evaluation framework with review and rapid debug templates.
- Improvement: the Changelog page now has its own SEO metadata (title, description, canonical) instead of inheriting the generic app title.
- Optimized prompt version updated to v1.1.6 (PROMPTEA: v1.1.6).
v1.1.5
Released: 2026-07-02
- New guide: Prompts for Gemini — format control with visible templates, grounded Q&A, and how to prevent verbosity.
- New guide: Prompts for Grok — how to get direct responses, using real-time context, and quick structured comparisons.
- New guide: AI prompt templates for business — project status reports, decision memos with criteria, and customer support templates.
- Improvement: Guides index page SEO description updated to reflect the full current guide catalog.
- Optimized prompt version updated to v1.1.5 (PROMPTEA: v1.1.5).
v1.1.4
Released: 2026-06-24
- New guide: Role prompting — how to assign expert personas for better outputs, with advisor and reviewer templates.
- New guide: ChatGPT prompts for work — professional email and executive summary templates with format constraints.
- New guide: Prompts for Claude — XML tags, extended context, and structured reasoning, with document Q&A and competing-viewpoints templates.
- Fix: the 'Start here' heading on the Guides index page is now properly localized in Spanish ('Por dónde empezar').
- Optimized prompt version updated to v1.1.4 (PROMPTEA: v1.1.4).
v1.1.3
Released: 2026-06-17
- Model registry updated with the latest OpenRouter models: GPT-5.5 and GPT-5.5 Pro, Claude Fable 5 and Opus 4.8, Gemini 3.5 Flash, Grok 4.3 and Grok Build 0.1, DeepSeek V4 Pro and Flash.
- New Perplexity provider (Sonar Pro and Sonar Reasoning Pro): full target support with engine hints, UI grouping, and prompt optimization tailored for real-time web search.
- Adaptive layer: the analyzer refines the optimized prompt with LLM.
- Fix: the 'Probar este prompt' / 'Try this prompt' button can no longer reset analyzer state while a result is visible, an analysis is pending, or files are loading.
- The optimized prompt header now shows v1.1.3 (PROMPTEA: v1.1.3).
v1.1.2
Released: 2026-05-31
- Persistent navigation: Promptea wordmark + links to Analyzer, Prompts, Guides, Models, and Glossary across all public pages.
- Reordered analyzer flow: prompt type is selected before writing, making attachment availability clear upfront.
- Live word and approximate token counter while writing the prompt.
- Example helper: button to prefill a sample prompt with a compatible type and model.
- Form state (prompt, type, model, format) is persisted in sessionStorage and restored on language switch.
- Focus trap in modals: Tab/Shift+Tab stay inside the dialog and focus returns to the trigger on close.
- Helpfulness feedback (Yes/No) now appears after the full analysis content, not before.
- Qualitative score label inside the quality ring: Weak (0–30), Fair (31–60), Good (61–85), Excellent (86–100).
- Improved optimized prompt: explanatory note about header lines, increased max height, and Copied! visual confirmation on copy.
- Count badges on prompt packs (e.g. 6 prompts) and Open Analyzer CTA on each pack subpage.
- How this works trigger redesigned as a more visible badge/pill for new users.
- Footer: | separators replaced with CSS gap, nav landmark, naturally formatted copyright text.
- Internal SEO indexing note removed from the public Guides page.
v1.1.1
Released: 2026-04-30
- More adaptive prompt analysis: output now changes depending on the selected model (GPT, Claude, Gemini, Grok, DeepSeek, Kimi) and shows model-specific notes.
- Centralized model registry with submodels (GPT-4.1 / GPT-4o / GPT-4o mini / o3, Sonnet / Opus / Haiku, Gemini Pro / Flash, Grok, DeepSeek, Kimi and Llama 3) and a submodel selector when the target supports it.
- More personalized scoring per use case: visible criteria for coding, debugging, refactor, data, marketing, study, image, research, and text.
- More actionable feedback: new “what works already”, “quick wins”, “helpful follow-up questions”, and “notes for this model” blocks in the results panel.
- Optimized prompt format toggle: Checklist (human-readable) or pure JSON (strict and parseable) with an explanatory tooltip.
- Improved attachment UX: clearer copy about supported formats, per-file size limits, and total caps.
- Optional Groq integration for extra hints — disabled by default. Set GROQ_API_KEY in Vercel to enable it; without the key, the app keeps using the local engine unchanged.
- Light/dark theme toggle re-enabled with system preference respected.
- New “Prompt of the Day” with deterministic per-date rotation: sticky aside on desktop, collapsible on mobile.
- Accessible “How this works” modal (ESC to close, initial focus, ARIA) explaining the flow step by step.
- New SEO landings: /landing/prompt-analyzer, /ai-prompt-optimizer, /prompt-scoring, /json-prompt-generator, /coding-prompt-generator, and /prompt-generator-for-{claude,chatgpt,gemini}.
- Refined sitemap, keywords, and metadata; canonical and per-language alternates updated.
- Additional tests: model registry, parseable JSON output, daily prompt rotation, and engine wiring of the format option.
v1.1.0
Released: 2026-03-10
- Refined engine with more task-aware behavior: scoring, findings, and recommendations now better distinguish between summarization, translation, code, data, study, marketing, and other use cases.
- New type filters: Translation and Summarization, with optimized prompts, output formats, and constraints tailored to each use case.
- Improved builder: optimized prompt version is now controlled via ENV, and output structure is more deeply tailored by purpose + task type + selected AI.
- Initial support for text/context attachments in compatible prompts: .txt, .md, .json, .csv, logs, and code files, used as real context in the analysis flow.
- New ATTACHED CONTEXT / CONTEXTO ADJUNTO block in the optimized prompt, explicitly treating files as context rather than instructions.
- Security improvements for attachments: strict validation of file types and sizes, content sanitization, and backend rejection for unsupported prompt types.
- Engine false positives fixed: it should no longer flag prompt injection because of Promptea’s own scaffold or lose context when re-analyzing optimized prompts with embedded files.
- More specific findings and recommendations for summarization and translation: instead of generic diagnostics, it now better detects missing focus, summary length, target language, or register.
- Expanded telemetry: beyond base analysis, the system now stores uploaded attachment format metadata (kind/ext/mime) for better product usage insights.
- Expanded test coverage for builder, engine, route, and attachment flows, including telemetry metadata persistence.
v1.0.3
Released: 2026-02-08
- SEO/Indexing improvements: reduced URL noise and consolidated canonical routes to avoid duplicates.
- Optimized hubs (Guides/Models/Glossary): added editorial content + internal links to improve crawl and indexing.
- Campaign tracking: added measurement tags (Google/Ads) to record usage conversions without exposing sensitive content.
v1.0.2
Released: 2026-02-03
- New SEO packs: Text and Data/JSON (copy-paste templates + deep links that prefill Promptea).
- New indexable sections: Guides, Models and Glossary (with per-slug pages and FAQ JSON-LD).
- Sitemap updated to include /prompts (incl. text/data), /guides, /models and /glossary + slugs.
- Prompts index updated to include the new packs (Text and Data).
- Footer reorganized: primary links on top and “Useful resources” on a separate line (Prompts/Guides/Models/Glossary).
- UI: pill styling updated (hover + selected) to transparent background + white border/text.
- Fix: analyzing Data/JSON prompts that were returning an error.
v1.0.1
Released: 2026-01-30
- SEO-friendly prompt packs (Study, Code, Marketing, Image) + deep links that prefill Promptea.
- FAQ per pack + Schema.org JSON-LD (FAQPage) to improve search CTR.
- UX improvements for navigation and faster iteration (quick access to templates).
- Footer updated with X link + tooltip.
v1.0.0
Released: 2026-01-27
- Modular engine + idempotency tests.
- Lint layer with stable IDs + output format detection.
- ES/EN calibration dataset to validate engine behavior.
- UI headline/bullets powered by scoreExplain + confidence badge.
- Firestore telemetry with TTL + user feedback.