Why bring-your-own-key

Most community platforms sell AI as a feature priced per seat or per token, with a fat markup on top of what the underlying model actually costs. We refuse to participate in that. AI capability is too useful to gate behind a metering layer, and the markup is the part of the deal we cannot defend. So mobieusAI is bring-your-own-key: tenant admins drop their Anthropic API key into the AI settings page, and every AI feature in the platform routes through it. The bill goes straight to your Anthropic account; you see exactly what each capability costs in dollars, in real time, on Anthropic's own dashboard.

What that buys you in practice. There is no upsell: either you have an Anthropic key or you do not. There is no per-seat charge, so a community of 50 and a community of 50,000 pay Anthropic the same per-token rate and nothing to us on top. And you can audit the spend on your Anthropic dashboard, with our settings page showing which capabilities are calling the model and at what volume.

Moderation AI: first-pass triage on every report

Reports arrive pre-categorized

When a member reports a post, the AI tags the report with a suggested category (spam, harassment, off-topic, doxing, hate, violence, sexual content, copyright) before a moderator sees it. The moderator can accept or change the category in one click, the AI is a draft, not a decision. Pattern detection runs across the report log: if the same offense type recurs three times in a week, the AI proposes a content rule that, once approved, auto-handles the next occurrence.

Draft replies that sound like your team

Every report gets a draft response, a polite, clear message to the reporter that explains the decision. Moderators edit before sending; the AI stays out of the final word. The draft pulls tone cues from your existing replies so it does not sound like a template.

Member AI grounded in your community

Members ask questions and get answers stitched together from your forums, mobieusKnow articles, and pinned threads, with citation links back to the source. Retrieval is full-text search against your own tenant database; there is no third-party vector store, no external indexing service, no copy of your community sitting on someone else's storage. The model does the synthesis; your community provides the ground truth.

This matters for two reasons. The answers stay current, when a forum thread updates, the next answer reflects it; no nightly re-index. And the data does not leave. The model call ships the retrieved snippets to Anthropic for the answer; nothing else.

Admin AI: analytics that explain themselves

The community analytics dashboard used to be charts you stared at. Now it has a one-line input next to each chart: ask "why did new-member signups drop on Tuesday" and the AI reads the metric, the period, the recent activity log, and the cohort breakdowns, and explains in plain language with links back to the underlying activity. Useful for community managers whose primary job is reporting upward to a sponsor or a board.

The Admin AI also drafts long-form summaries on demand: a weekly digest of forum activity, a month-end engagement narrative, a quarterly health report. Each draft is sourced, every claim links to the metric that supports it, so the human editor can verify before it goes out.

What is on the BYOK roadmap

The model gateway is portable. Future releases will let tenants point mobieusAI at OpenAI, Google Vertex, or a self-hosted model with the same UI and the same prompts. Today we ship Anthropic because Claude Sonnet 4.6 is what we use ourselves to build the platform and we trust its behavior end-to-end. When you bring your own key, you opt into a model lineage you can verify.

How it interacts with your other Mobieus surfaces

  • mobieusHelp ships eight AI hooks with the helpdesk: reply suggestions, ticket summary, sentiment scoring, auto-categorize, resolution prediction, knowledge-base gap detection, audit Q&A, canned-response generation. Each hook is gated on the AI master switch; turn AI off and the hooks vanish.
  • mobieusLearn gets quiz generation, lesson takeaways, assignment rubrics, and course descriptions. The author writes the spine; the AI fills in scaffolding that the author edits.
  • Chat rooms get a topic-suggest tool on the room-edit page, and the moderation surface scores chat messages for hate or harassment in real time.
  • Public API exposes every AI capability as a first-class REST endpoint on Pro and up, so your own tooling can call them too.

What it costs to run

This is up to you and Anthropic. Claude Sonnet 4.6 is priced at Anthropic's published rate; the bill scales with how active your community is and which capabilities you enable. We do not insert a metering layer between you and the model; the costs are theirs to publish and yours to monitor.

Your bill scales with how active the community is and which capabilities you turn on. The settings page tracks token volume per capability so you can see what is pulling weight, throttle anything that is not earning its keep, and forecast spend against your Anthropic dashboard. Light moderation use on a small community is a different month from a high-traffic forum with admin narration on every metric. For how the model you pick changes that math, see AI cost control by tenant.

Get an Anthropic key, drop it in

Anthropic onboards keys at console.anthropic.com. Pick a workspace, generate a key, paste it into the AI settings page on your tenant. Every AI capability lights up immediately. Disable any individual capability you do not want; turn the master switch off and no AI calls leave your tenant.

Frequently asked questions

Do I need an Anthropic key to use Mobieus?

No. Every Mobieus feature works without AI. mobieusAI is an accelerator on top: add an Anthropic key when you want moderation triage, the member assistant, and analytics narration, and skip it if you do not.

Does my community data get sent to Anthropic?

Only the snippets needed to answer a given prompt. Retrieval runs as full-text search against your own tenant database, with no third-party vector store and no external index. The model call ships those retrieved snippets to Anthropic and nothing else.

Can I turn AI off, or limit which features use it?

Yes. The AI settings page has a master switch plus a per-capability toggle. Disable any single capability, or flip the master switch and no AI calls leave your tenant.

What does mobieusAI cost?

Whatever Anthropic charges for the tokens you use, billed straight to your Anthropic account with no Mobieus markup and no per-seat fee. The settings page tracks token volume per capability so you can forecast spend.