New Bring your own LanguageTool, Firecrawl and search keys

Articles you can check sentence by sentence

Pennao finds scholarly publications and trusted pages, writes to your keyword and checks the text before you see it. You get a finished article with source links on the exact words, metadata and JSON-LD.

Free to use. You pay your model provider directly for tokens.

The animation shows how Pennao prepares an article: it finds and scores sources, writes text with links to them, removes phrases typical of generated text and gives the article a quality score.

Works with the models you have keys for

  • OpenAI
  • Anthropic
  • Google Gemini
  • OpenRouter
  • Kimi
  • Mistral
  • DeepSeek
  • Qwen
  • xAI

Pennao in numbers

articles generated
20,000+
users in 310 teams
1,150+
sources fetched and scored
412,000+
average quality score
91/100

From the dashboard, as of November 2026.

From brief to publication

Six stages, each visible in the dashboard

Follow progress live and go back to any stage afterwards: the sources found, the plan, the check results and the cost of every model call.

  1. 01

    Brief

    Keyword, intent, audience and the links that must appear in the text. Your site profile fills in the rest: offer, tone and pages to link to.

  2. 02

    Research

    OpenAlex, Semantic Scholar, Crossref, PubMed and pages from the web. Every source gets a score from 0 to 100: type, citations, recency and relevance.

  3. 03

    Plan

    An analysis of the search results and a heading plan for the keyword. A second model reviews the plan and points out gaps. You can approve it before writing starts.

  4. 04

    Writing

    Sections are written in parallel, each aware of its neighbours. The link to a source lands on the words the information belongs to.

  5. 05

    Quality check

    LanguageTool, fact-checking, the AI-slop filter and an SEO linter. Text below the threshold goes back for fixes, up to three times.

  6. 06

    Publication

    Markdown, HTML or JSON with JSON-LD, WebP images in several sizes and translations with hreflang. Then a webhook to your CMS.

Features

Everything that happens to the text after the first draft

Pennao stays with the article after it is generated: through review, translations and later versions.

A source for every claim

The link goes from the exact words to the page or paper the information comes from. A sentence without a source goes back for a fix or is removed.

A quality bar you set yourself

A score from 0 to 100 built from language, style, readability, SEO and links. Text below your threshold does not reach you until it passes the fixes.

AI-slop filter

Removes empty openers, piles of adjectives and formulaic summaries. It has separate dictionaries for English, Polish, German, French and Spanish.

Edit one passage at a time

Select a paragraph and ask for a shorter version, another angle or an extra source. You review the changes in a diff and accept them one by one.

10 languages from one order

Translations with their own keywords and hreflang tags. Every version goes through the same quality check.

  • EN
  • PL
  • DE
  • FR
  • ES
  • IT
  • PT
  • NL
  • CS
  • UK

Model costs under control

Cost and tokens of every call, daily and monthly budgets with a hard stop. When a model hits the provider limit, the stage switches to a fallback model and logs it on the timeline.

Team and permissions

Owner, admin, copywriter, viewer and custom roles. Every change shows who ordered it and who edited the text.

Style memory and authors

Your edits turn into style rules that you approve before they are used. Author personas have their own voice, rules and expertise.

Topic ideas every week

An agent reads your site and suggests topics every week from Google Trends, news, new publications and gaps in the project knowledge base.

Ready to publish

Markdown with frontmatter, HTML or JSON with blocks, a table of contents and JSON-LD (BlogPosting, FAQPage, BreadcrumbList). WebP images in several sizes.

Quality check

Six checks before the text reaches you

Pick a check to see what it changed in this paragraph. The full rules are on the Quality page.

art_01K7… 92/100

It is worth noting that without a battery, the grid takes 60–70% of the output, because nobody is home during the day. A study of 312 households3 found that a battery raises self-consumption to 65–75%. At €0.30 per kWh, a home battery with solar pays off faster, because energy used on site is worth nearly twice as much as energy sold. Battery prices keep falling, so the same kit may cost less in two years.

API and webhooks

Order an article with one request

You get a ticket, live progress over SSE and a webhook signed with HMAC when the text is ready. The same flow works from the dashboard, the API and the MCP server.

  • Scoped tokens limited to the projects you choose
  • Idempotency-Key: a retried request never creates a duplicate
  • Live events over Server-Sent Events
  • HMAC-SHA256 signature and up to 8 delivery attempts per webhook
Developer documentation

order.sh

curl -X POST "$PENNAO_URL/v1/articles" \
  -H "Authorization: Bearer $PENNAO_TOKEN" \
  -H "Idempotency-Key: battery-2026-11-17" \
  -H "Content-Type: application/json" \
  -d '{
    "projectId": "prj_01K7…",
    "brief": {
      "title": "Is a home battery worth it with solar?",
      "keywords": { "primary": "home battery with solar" },
      "intent": "commercial"
    },
    "settings": { "languages": ["en", "de", "nl"], "includeFaq": true }
  }'

# 202 Accepted
{ "ticket": "art_01K7…", "status": "queued" }
Keys and costs

You pay model providers directly, with no markup

Pennao runs on your organization’s keys. Using Pennao is free, and you settle tokens with the provider at its own prices.

  • Keys are encrypted with AES-256-GCM. After saving, only their last characters are shown.
  • Connect your own LanguageTool, Firecrawl and search engines, including servers you host yourself.
  • Budgets with a hard stop pause jobs before they go over the cost you planned.
How we protect keys and data
Questions

Frequently asked questions

Is Pennao free?

Yes. Using Pennao costs nothing. You pay your model provider (OpenAI, Anthropic, Google, OpenRouter and others) for the tokens the articles use, through your own API key.

Do I need my own API keys?

Yes, for the language models. You add an OpenAI, Anthropic, Google, OpenRouter or other key in your organization settings and can remove it at any time. LanguageTool, page fetching and the scholarly databases work without extra keys.

How much does one article cost in tokens?

A 1,500-word article usually costs from a few cents on budget models to a dollar or two on the strongest ones. The dashboard shows the exact cost of every stage under Usage and costs.

Where do the sources come from?

From scholarly databases (OpenAlex, Semantic Scholar, Crossref, PubMed, Europe PMC, arXiv, CORE, DOAJ) and from web pages. You can set preferred and blocked domains, and links that must appear in the text.

Will the text read like generated content?

The AI-slop filter removes phrases typical of generators in five languages, and the editor evens out sentence rhythm and length. Style memory built from your edits and author personas with their own voice do the rest. Read the text before publishing anyway: that is what the review view is for.

Does Google rank AI-assisted content lower?

According to Google Search guidance, what counts is how helpful the content is, not how it was produced. Sites lose out on text mass-produced without value for the reader. Pennao writes for a specific intent and backs every claim with a source.

How do I get articles into my CMS?

Download Markdown, HTML or JSON from the dashboard or the API. When an article is finished, Pennao sends an article.completed webhook, so you can automate publishing to WordPress, Webflow, Strapi or your own system.

Who owns the generated text?

You do. You can publish, edit and sell it without limits. Pennao does not keep archives or backups of your content, so export what you want to keep. Details are in the terms of service.

Try it on your own topic

Enter a keyword, add a model key and read a sourced article in about fifteen minutes.