The architecture that makes 75 features manageable.
Every one of the 75 features in mobieusAI routes through one Anthropic API key: yours. You drop it into your AI settings. The bill goes to Anthropic directly at their published rates, with no Mobieus markup on top. Your spend dashboard shows exactly what each feature has cost you this month in dollars, not tokens — a number you can actually act on.
Each feature has its own on/off toggle, a monthly spending cap, and a per-user daily call limit. You can enable the helpdesk reply drafts and the thread summarizer and nothing else. You can cap any member-facing feature at $10 a month and it stops firing when it hits $10. You choose the Claude model per tenant — Haiku, Sonnet, or Opus — and can override it per feature, so the cheap, fast model handles the high-volume member-facing calls and the better reasoner handles the judgment calls. The full cost-control breakdown is in the model selection post.
The result: 75 features that you can treat as one system or tune individually, depending on how much you care to manage them.
What members see.
Nine features live in the forum surface alone. When a member opens a new thread, the title suggester proposes three alternatives based on what they've drafted — one click to accept, dismiss, or ignore. Before they pick a category, "Where should I post?" reads the draft and surfaces the right forum. When they're composing a reply they want to soften or sharpen, the tone-shifter rewrites it. A long thread they want to catch up on quickly gets collapsed to key points by the thread summarizer, sourced from the actual posts.
The same logic extends to messaging. DM smart compose helps draft an opening message. Tone shift adjusts it. A translation hint surfaces when a member suspects a language gap. On the profile page, bio writing assist helps a new member say something real about themselves; the interest tag suggester then proposes tags from the bio they just wrote.
In mobieusMarket, a member listing a product gets description writing assist and title rewriting when the listing goes live. When they respond to a buyer inquiry, the reply tone helper keeps the exchange professional without requiring a second draft.
In mobieusHelp, before a member even submits a ticket, the knowledge base deflection feature reads what they're typing and surfaces the most relevant existing articles. It does not prevent submission — it gives them a chance to find the answer first.
Across all of this, the flagship mobieusAI Q&A assistant answers questions grounded in your community's forums and knowledge base. A member asking about your community's rules, your course materials, your posted FAQ — the assistant draws from what your community has actually produced, not from generic web knowledge. The same assistant runs at three scope levels: member, moderator, and admin, each with access appropriate to the role.
Members also get a daily briefing on the dashboard — a one-paragraph summary of what happened since they were last active, with links to the threads worth catching up on. A following catch-up surfaces the same, scoped to the people and topics they explicitly follow. Auto-organize bookmarks tidies their saved content into categories. Finish my draft reopens an abandoned post and helps them get it done. Reconnect icebreaker prompts a message to a member contact they haven't talked to in a while.
What moderators see.
When a report lands in the mod queue, it arrives auto-categorized before a human reads it. Bulk-moderation triage orders a batch of flagged items by urgency, with a short summary of each. Mod response drafts write a contextual reply to the reporter, citing the rules relevant to the report. Moderators approve, edit, or reject — nothing sends without a human decision.
Five forum tools let mods improve any thread they're looking at: title polish, body rewriter, excerpt generator, tag suggester, SEO meta-description. These are mod-initiated. They don't touch threads automatically.
In mobieusHelp, the moderator (support agent) surface is where AI does some of its densest work. Incoming tickets arrive pre-tagged and pre-categorized. Sentiment badges each one. The suggested canned reply ranker orders the existing response templates by fit rather than alphabetically. The reply draft writer generates a contextual first draft for the agent. Resolution prediction gives an estimate of how this ticket type typically closes, based on queue history.
What admins see.
Eight admin-facing features cover the administrative layer. The weekly insights digest narrates what happened in the community and flags where attention is needed. The audit log Q&A answers natural-language questions about past admin actions — "which admin changed the forum settings last Tuesday" works, with an answer. The config recommender reads your current settings and proposes adjustments. The broadcast composer drafts mass messages. The SEO description writer handles the tenant's meta fields. The content rule creator converts observed content patterns into formal moderation rules. The canned response generator builds help desk templates from scratch. The digest intro polisher tweaks the opening of your weekly member digest to match your voice.
The AI Community Manager — described fully in its own post — sits on top of all this. It runs proactively, drafting welcomes, weekly summaries, churn re-engagement DMs, member pair-up suggestions, forum reply drafts, and moderation assists into a queue at /admin/community-manager. Nothing sends until an admin approves. That constraint is structural, not a setting — there is exactly one code path that flips a draft from pending to sent, and it requires an authenticated admin session.
The course creator layer in mobieusLearn.
Building a course in mobieusLearn involves writing that most subject-matter experts find tedious: course descriptions, lesson takeaways, quiz questions, assignment rubrics. Nine features cover this layer.
The course description writer generates a student-facing description from the outline you've already built. The lesson takeaways tool extracts three to five key points from the lesson text. The quiz generator creates multiple-choice questions from lesson content — a first pass you edit, not a final product you ship blind. The rubric suggester proposes grading criteria from an assignment prompt.
For learners, there are three more: a peer comparison coaching nudge that surfaces when a member's progress falls behind their cohort median, an AI-written cohort weekly digest that summarizes the week's learning activity, and live session description assist for instructors writing the brief before a session goes live.
Two features face technical administrators: an xAPI failure explainer that interprets outbound statement errors in plain English instead of raw error codes, and a SCIM audit anomaly summary that flags unusual provisioning patterns for review.
The knowledge layer in mobieusKnow.
Five features in mobieusKnow. An outline generator scaffolds a new article from its title. The title suggester proposes clear names when someone has a draft body but no title. A clarity rewrite simplifies dense text without changing the meaning. The stale-page detector flags articles that haven't been touched while the underlying topic has moved on — it catches documentation debt before it misleads members. Suggested wiki edits give moderators reviewing recent changes a starting point for corrections.
The helpdesk layer in mobieusHelp.
mobieusHelp has the highest concentration of AI assists in any single product: eight. Incoming tickets get auto-tagged and auto-categorized before a human touches them. Sentiment analysis badges each ticket. The canned response generator builds new templates on request. Reply suggestions are ranked by fit. The draft writer produces a first-pass response for the agent. KB gap detection runs on a schedule and surfaces topics where members submitted tickets but the knowledge base has no article — a direct signal for what to write next. Resolution prediction tells agents how ticket types like this one typically close.
Taken together, the help desk AI does the work that usually sits between "ticket assigned" and "agent starts typing" — pre-reading, pre-sorting, pre-drafting — so the agent's decision is the focus, not the setup.
Seventy-five, and growing.
The count sits at 75 today. The list is growing as gaps surface from real usage — the shape of the system (AI woven into the moment of action, per-feature toggles, one key) is settled, but the number will keep moving. If you want to see every feature with its current status, the full grid lives at /product/mobieusai.
Frequently asked questions.
Do I need a separate AI subscription beyond Mobieus?
Yes. You need an Anthropic API key. It's separate from your Mobieus subscription. You sign up at Anthropic, generate a key, paste it into your AI settings page, and Anthropic bills you directly at their published rates. There is no Mobieus charge on top.
Can I turn on only some of the 75 features?
Yes. Every feature has its own toggle. You can enable the help desk reply drafts and the forum thread summarizer while leaving everything else off. You can also set a per-feature monthly spending cap and a per-user daily call limit — so a high-volume member-facing feature can have a hard ceiling that stops it from running away.
Does mobieusAI have access to my members' private data?
The Q&A assistant is grounded in content visible to the member's role. Member-scope Q&A sees what members see. Mod-scope sees what mods see. Admin-scope sees what admins see. Private messages are not used as training material or as grounding context. The model itself is Anthropic's; your key governs what you send to it, and the platform scopes what data each role is allowed to include in a query.