Getting started

A 10-minute walkthrough: install the free edition for your CMS, configure a provider, send your first chat turn, then add Pro when you want the agent to build for you.

1. Install the free edition

WordPress (5 min)

Requires WordPress 6.4+ (tested up to 6.9, multisite supported) and PHP 8.1+. WordPress 6.3 and earlier are unsupported.
  1. Plugins → Add New in the admin.
  2. Upload Plugin and pick seanosai-1.0.0.zip from

/products/seanosai-wordpress (or, once we're listed, search "Seanos AI" on WordPress.org).

  1. Activate. A new top-level menu Seanos AI appears with

submenus: Dashboard, Conversations, Tools, Reporting, Settings.

  1. Visit any of those once so the plugin runs its first-time database

setup.

Joomla 4.2+ / 5.x (5 min)

Requires Joomla 4.2+ or 5.x and PHP 8.1+. Joomla 3.x and Joomla 4.0 / 4.1 are unsupported (the codebase uses J4-only patterns and requires PHP 8.1, which Joomla 4.2 was the first to mandate).
  1. System → Install → Extensions → Upload Package File.
  2. Upload com_seanosai-1.0.0.zip from

/products/seanosai-joomla (or, once we're listed, install from JED).

  1. Components → Seanos AI lands you on the Dashboard. Submenus

appear in Joomla's left admin sidebar: Dashboard, Conversations, Reporting, Tools, Settings.

  1. Reload the admin once so the install script seeds the chat /

conversation tables.

The free edition works on its own from this point — no API key, no remote service. The next step adds an LLM provider so the agent can actually answer.

2. Configure a provider

Seanos AI → Settings.

Pick whichever provider you have a key for:

ProviderWhere to get a keyNotes
Anthropic (Claude)console.anthropic.comBest tool-use reasoning. Free tier sufficient for casual use.
OpenAIplatform.openai.comWidest model selection.
Geminiaistudio.google.comFree tier, fast.
DeepSeekplatform.deepseek.comCheap. Good code-aware reasoning.
Grok (xAI)console.x.aiStrong long-context and reasoning.
GitHub Modelsgithub.com/marketplace/modelsFree for personal use.
Ollama (local)install OllamaRuns on your hardware. No data leaves the box.
OpenAI-compatibleany URLSelf-hosted vLLM, LM Studio, LocalAI, etc.

Toggle the provider on, paste the API key, click Test connection. A successful test also persists the key — no separate Save click needed. The plugin / component encrypts the key at rest with sodium-secretbox keyed off the platform's site secret (AUTH_KEY on WordPress, the Joomla secret on Joomla); it's never returned to the browser, and the Settings UI shows Key saved once configured.

3. Pick the model in the chat header (per conversation)

There is no global default model any more. Each conversation has its own provider / model selector in a compact chat header — pick a chat provider, a chat model, and (optionally) a separate image provider and image model for generate_image calls. Defaults are sensible (Claude Sonnet on Anthropic, GPT-4o-mini on OpenAI, Gemini Flash on Gemini), and the first option per provider is Default, which auto-tracks whatever the latest model is for that provider.

You can chat with one provider and generate images with another in the same conversation — for example, chat with Anthropic, generate images with OpenAI's gpt-image-2. If you don't pick an image provider, the tool falls back to the chat provider (which only works with OpenAI or Gemini today; other providers return a "switch provider" error).

Click Test connection in Settings to verify each enabled key round-trips before you commit to it.

4. Send your first chat turn

Dashboard has the curated prompt library — every card is a multi-step prompt the agent will follow once you click Start.

A safe first run on either platform:

  1. Click the Site health summary card (free tier, read-only tools

only).

  1. The agent inspects the install, lists active extensions, reads

the global config, and writes back a one-screen overview. Nothing is written to disk or the database.

Or click Blank chat in the top-right and just type. Try:

List my logs, then tail the most recent one and pick out anything that looks like a recurring error.

4.5. Try image generation

If you have an OpenAI or Gemini key configured, every conversation can generate images. In the chat header, set the image provider to OpenAI (DALL·E 3 / gpt-image-1 / gpt-image-2) or Gemini (Imagen 3 / gemini-2.5-flash-image / "Nano Banana"), then ask:

Generate a header image for a blog post about beekeeping — soft watercolor style, no text.

The agent calls generate_image, the result is saved into the media library (wp-content/uploads/seanosai-ai/ on WordPress, /images/seanosai-ai/ on Joomla), and the chat renders it inline. Like every write tool, it pauses for approval the first time it runs. Image generation bills against the configured provider's API; see the FAQ for cost notes.

Other providers (Anthropic, Grok, DeepSeek, GitHub Models, Ollama, generic OpenAI-compatible) don't implement image generation and return a "switch provider" error if you try.

5. Drop attachments into chat

You can drop files into the chat composer alongside your prompt. Four buckets are accepted:

  • Images (.png / .jpg / .gif / .webp) — the model sees the

image natively if it's vision-capable; text-only models return a graceful error and you can switch model.

  • PDFs — Anthropic and Gemini read them natively (tables, layout,

embedded images intact). On OpenAI / Grok / DeepSeek / GitHub Models / Ollama / generic OpenAI-compatible the plugin shells out to pdftotext (poppler-utils) for plain-text extraction; install poppler-utils on your server, or switch to Anthropic / Gemini for full-fidelity PDF reading.

  • Office files (.docx / .xlsx / .pptx) — extracted to plain

text server-side via PHP's native ZipArchive + DOMXPath and inlined. Legacy .doc / .xls / .ppt aren't supported — re-save as OOXML. 10 MB cap per file.

  • Plain text (.txt / .md / .csv / .json / source files) —

inlined as a text block, capped at 200 KB per attachment.

Every provider sees attachments now (it used to be Anthropic + Gemini only).

6. Approval mode

Every conversation starts in strict approval mode. When the agent calls a write tool (create_article, disable_extension, etc.), it pauses with an inline approval card:

awaiting approval — create_article [Approve] [Reject]

Approve runs the tool. Reject writes a synthetic "rejected" tool result and the model continues without the side effect. You can flip a conversation to trusted mode (Conversation page → top-right) to run writes immediately — useful once you trust a particular prompt or workflow.

7. Install Pro

Want the agent to actually build things for you? File edits, plugin / module / template / component scaffolding, write SQL, PHP lint, package builds, snapshot rollback?

WordPress

  1. Buy Seanos AI Pro for WordPress

— $49, lifetime updates.

  1. Download seanosai-pro-1.0.0.zip from your receipt email.
  2. Plugins → Add New → Upload Plugin → upload the zip → Activate.
  3. The Tools page now shows 16 new entries; the Dashboard shows the

advanced prompt cards as startable (they were locked with a "Get Pro" button before).

Joomla

  1. Buy Seanos AI Pro for Joomla —

$49, lifetime updates.

  1. Download plg_system_seanosaipro-1.0.0.zip from your receipt

email.

  1. System → Install → Extensions → Upload Package File → upload

the zip.

  1. Extensions → Plugins → enable Seanos AI Pro. The Tools

page picks up the 16 new entries on the next request.

Pro registers itself on the free edition's public extension hook (seanosai_register_tools action on WordPress; onSeanosaiRegisterTools event on Joomla), so the free edition picks it up the moment Pro activates. No setting to flip, no key to paste.

What's next

gated, and which platform it runs on.

multi-step prompts.

  • FAQ — common gotchas.