Since Notifuse 22, the blog editor and the email template editor each have an AI assistant. You open a chat beside the editor, say what you want, and the assistant writes it into the editor, whether that's a whole blog post with its title and SEO fields or an email template built block by block. It runs on your own API key, with Anthropic, OpenAI or Gemini as the provider. Notifuse doesn't resell AI credits. Your server calls the provider directly, and the provider bills you.
The assistants
The AI Blog Assistant came with Notifuse 22.0, in the blog post editor, and the AI Email Designer followed a day later in 22.1, in the email template editor. Each opens from a round button in the bottom-right corner of its editor. Since 38.0 there is a third one, the Analytics Assistant, which uses the same integration to answer questions about your traffic in Web Analytics. This post covers the first two.
Before every message, the assistant reads what is on screen: for a blog post, the text, title, excerpt, SEO and Open Graph fields; for an email, the template's block tree, subject line, preview text and category. So you can say "shorten the intro" or "make the button bigger" without pasting anything into the chat.
Blog posts
The blog assistant has two tools. One writes the post body into the editor as rich text, with headings, paragraphs, lists, quotes, code blocks, links, images and YouTube embeds. The other fills the metadata: title, excerpt, meta title, meta description, keywords, and the Open Graph title and description. You can ask things like:
"Write a 900-word post announcing our new pricing, with one section per plan."
"Write the excerpt, an SEO title and a meta description under 160 characters."
Writing the body replaces the whole document in the editor, but nothing is saved until you save the post yourself.
Email templates
The email assistant works on the same block tree as the visual editor, so you can still edit by hand anything it builds. It can add, update, delete and move blocks (sections, columns, text, buttons, images, dividers, spacers, social icons), or replace the whole tree to start from scratch. It selects a block to show you what it changed, and it sets the subject line and preview text, Liquid included. Since Notifuse 44, it can also give a block a display condition, so that it shows only to some recipients, such as "order.total is greater than 100". You can ask things like:
"Design a product launch email: a hero image, three feature columns and a call to action."
"Make the button green and center it."
"Write a subject line that greets the contact with
{{ contact.first_name }}."
The assistant designs for the category set in the template's Settings tab. A marketing template ends with a footer that holds a visible unsubscribe link and a link to the preferences page, and the assistant keeps that footer when it rebuilds the layout. A transactional template gets no unsubscribe link, since an API send has no list. A double opt-in template gets its confirmation button. The assistant also knows which Liquid variables each kind of send fills.
Before a template rebuilt from scratch replaces yours, Notifuse checks it against the MJML nesting rules (a column must sit in a section, a button in a column, and an image needs a source). If the check fails, the chat shows what is wrong and your template stays as it was. After each edit the assistant makes, Notifuse compiles the template, and if it doesn't render, the chat lists the MJML errors so you can ask for a fix. As with blog posts, nothing is saved until you save the template, and since Notifuse 44, Undo in the visual editor takes back a change the assistant made.
Web search
If you add a Firecrawl integration, the assistants get two more tools, which run on your Notifuse server. search_web searches the web and returns links with titles and descriptions. scrape_url reads a page and returns its main content as markdown. The chat shows each search and page read as it happens. With these you can ask:
"Read our changelog at https://example.com/changelog and write a post about the last three releases."
"Search for recent articles on Gmail's bulk sender rules and sum up what changed."
One message can chain up to ten rounds of searches and reads before the assistant answers. Firecrawl is optional and bills you directly. You can also point it at a self-hosted Firecrawl, since its form takes a base URL.
Providers
As a workspace owner, you add a provider in Settings → Integrations with your API key and a model:
Anthropic takes a suggested model or, since 44.0, any Anthropic model ID you type.
Gemini, available since 34.0, takes the suggested model or any Gemini model ID you type.
OpenAI, available since 29.2, takes any model name. Its optional Base URL field points the integration at any OpenAI-compatible endpoint, such as OpenRouter, which reaches most models in the world with one key, or a model you run yourself. The Reasoning effort setting, added in 35.0, controls how much reasoning models think.
If you have more than one integration, you pick one from a menu at the top of the chat. Notifuse encrypts every key, the Firecrawl one included, with your instance's secret key before storing it.
In the chat
A model that reasons before answering streams its thinking into a collapsible Thinking panel. The assistants make every change in one response and build the article or email before the title or subject. If the token limit cuts a reply off, the changes are already in the editor, only the commentary is lost, and the chat shows a warning. When a model makes its changes without writing a word, Notifuse sends it the results once more so it can reply.
With Anthropic and Gemini, the bottom of the chat shows what the conversation has cost so far, split into input and output, from the token counts the provider returns and Notifuse's own price list (a model missing from it counts as zero). OpenAI integrations show no cost, since a compatible endpoint can serve any model at any price. New conversation clears the thread and resets the counter.
Using an assistant takes the LLM write permission, which full access includes. Since it spends your provider credits, think of that permission as a spending grant.
Not yet
Conversations aren't saved. The thread lives in your browser while the editor is open, and nothing is stored on the server.
The assistant doesn't review its own result. Its edits land in the editor once it has answered, and it only sees the updated post or template with your next message.
A reply is capped at 4,096 output tokens in the blog editor and 8,192 in the email editor, thinking included, so write a long article section by section.
The email assistant only works in the visual editor. It is hidden while a template is in code mode.
Getting started
If you self-host, run Notifuse 44.0 or later. As a workspace owner, open Settings → Integrations, add Anthropic, OpenAI or Gemini with your API key and a model, and add Firecrawl if you want web search. Then open a blog post or an email template and click the assistant's button. Until an AI integration exists, the button opens a short panel with a link to the integration settings.
The AI assistant documentation covers the blog and email assistants, and the Analytics Assistant has its own page. The Anthropic, OpenAI, Gemini and Firecrawl pages walk through each integration.
