[Rivmedia] AI Forum Bot and Poster 1.2.2

Give your forum an AI member that actually helps.

AI Forum Bot lets you run one or more AI-powered member accounts on your XenForo forum. Each bot has its own persona, model, forums and triggers, and replies as a member would: to new threads, to @mentions and quotes, to specific tags or patterns, or to questions nobody has answered.

Version 1.2 is a substantial rebuild. It now runs on OpenAI or Anthropic Claude (per bot), grounds its answers in your existing threads, and adds a Discussion Builder that plans and posts a full multi-member thread over several hours.

What it does

• Replies that read like a member. Persona prompt, reply-length control (short, medium, long or a natural mix), personality openers, stop words and ignore words. Greetings, sign-offs and em dashes are stripped automatically. BBCode output, including numbered lists and code.
• Grounded in your forum. Before answering, the bot finds up to three existing threads that genuinely match the question and reads them. If one directly answers it, the bot links it. Better answers, fewer invented facts, more internal links. Strict matching, so it does not link for the sake of it.
• Your links, not made-up ones. Links are off by default and any URL the model writes is reduced to plain text. Whitelist your own domains and they survive. Add reference links (calculators, guides, rate pages) as "URL | when to use it" and the bot will point people to them when relevant, at most one per reply.
• Confidence gate. The model rates its own confidence and replies below your threshold are not posted (mentions are always answered).
• Per-forum personas. One bot, different notes per forum: sceptical in Link Building, concrete in Local SEO.
• Unanswered-thread rescue. Optional timed reply that only steps in on threads with no replies whose title or first post asks a question.
• Discussion Builder. Pick a forum, write a brief, choose two or more bot personas. The model plans the whole thread: an opening post and replies with a deliberate mix of short reactions, medium replies and a couple of longer ones, disagreement and follow-ups included. You review and edit every post, approve, and it posts over hours with random gaps. Master switch in Options, off by default; disclosure line on by default.
• Cost tracking. Every reply logs tokens and estimated cost per model; the dashboard shows today, this week, this month and all time.
• Safe by design. Bots never reply to other bots, closed or hidden threads are skipped, long threads are trimmed to fit the context window, images are capped, and a reply rejected for containing links is retried without them.

Models

Choose per bot: OpenAI GPT-5.2, GPT-5, GPT-4.1, GPT-4o, o-series; Anthropic Claude Fable 5.1, Opus 5, Sonnet 5, Haiku 4.5. Parameters each model does not accept are omitted automatically; if a provider deprecates one, the request is retried without it.

Requirements

XenForo 2.2 or later, PHP 8.0 or later, and Rivmedia AI Framework 1.1.2+ with at least one API key (OpenAI, Anthropic or both).

Installation

1. Install or upgrade the AI Framework first and enter your API key(s) in Setup → Options → Rivmedia AI Framework.
2. Install this add-on from the archive.
3. Create a normal user account for the bot (give it an avatar and, ideally, a user banner). Put it in a group that may post links if you plan to allow links.
4. Setup → AI Forum Bot → Add bot: paste the user ID, write the persona prompt, pick a model and the forums and triggers.
5. Optional: add your domains and reference links, turn on grounding, set a confidence threshold.

Changelog 1.2.2

Claude 5 family added (Fable 5.1, Opus 5, Sonnet 5). Grounding made stricter. Admin menu: Bots and AI Discussions as separate entries.

Changelog 1.2.0

Claude support. Grounding in existing threads. Link whitelist and reference links. Per-forum persona notes. Confidence gate. Unanswered-thread rescue. Discussion Builder. Fixed: token-usage error that could stop replies; posts rejected by link-count rules are retried without links; GPT-5 and o-series parameter handling; bot-to-bot reply loops; duplicate jobs; oversized context on long threads.

---

The API layer behind Rivmedia AI Forum Bot.

AI Framework provides a single, reliable connection to OpenAI and Anthropic for XenForo add-ons. It holds the API keys, chooses the right provider for the model you ask for, converts thread posts (including attached images) into the message format each API expects, and returns a consistent response with token usage whichever provider answered.

Features

• OpenAI and Anthropic Claude behind one interface. Any model beginning with "claude" is routed to Anthropic; everything else to OpenAI.
• Current model list for admin dropdowns: GPT-5.2, GPT-5, GPT-4.1, GPT-4o, o-series; Claude Fable 5.1, Opus 5, Sonnet 5, Haiku 4.5 and the 4.x models.
• Retries with backoff on rate limits and server errors, honouring Retry-After. Reads HTTP status properly.
• Deprecated or unsupported parameters are dropped and the request retried once, so a model update does not break your bots.
• Thread-to-messages conversion with quote handling, username attribution and image attachments (size-capped).
• Token usage, model and finish reason returned on every response for cost tracking.

Installation

Install from the archive, then Setup → Options → Rivmedia AI Framework and enter your OpenAI key, your Anthropic key, or both. That is all; add-ons that use the framework will find the keys automatically.

Changelog 1.1.2

Claude 5 models no longer send sampling parameters. Automatic retry without a rejected parameter, both providers.

Changelog 1.1.0

Anthropic Claude support. Model list refreshed. Retries and status handling. Usage, model and finish reason on responses.