> ## Documentation Index > Fetch the complete documentation index at: https://collony.ai/docs/llms.txt > Use this file to discover all available pages before exploring further. # Bot Setup > Configure the content and rules the collony agent uses to moderate and engage your community. Bot Setup is where you define what the agent knows and how it behaves. All changes are versioned - you can see who made the last edit and revert if needed. Bot Setup 1 ## Project About A short 2-5 sentence summary of your project or community. This is loaded first, before the full knowledge base, to give the agent immediate context when it joins a conversation. Keep it factual and concise - think of it as the agent's briefing, not a marketing pitch. **Example:** "We're a platform that helps indie game developers publish and monetize their games. Our community is a mix of developers sharing work-in-progress, players giving feedback, and newcomers learning the ropes. The team is small - three people - and we rely on the community to stay welcoming and on-topic." ## Knowledge Base The knowledge base is what powers the AI chat function. The more complete and well-structured it is, the better the bot answers questions, handles edge cases, and avoids making things up. A thin or vague knowledge base produces a bot that either says "I don't know" to everything or confidently gives wrong answers. **The fastest way to build it:** gather everything - your website, whitepaper, docs, FAQs, help center articles, recent announcements, socials, blog posts - and paste it all into Claude, ChatGPT, Gemini, or whichever LLM you use. Ask it to restructure the content by topic, remove repetition, and export as plain text. Then paste the result here. A complete knowledge base covers: * What the project is, how it works, and what problems it solves * Every question your community asks regularly, with clear answers * Official links, social handles, and where to find what * Key policies - what's allowed, what's not, how support works * Team contact handles so the bot knows who to tag when something falls outside its scope * Scam awareness reminders - who the team will and won't contact members through The AI chat function is only as good as what's in the knowledge base. If something isn't in here, the bot will either guess or decline to answer. Keep it updated as your project evolves. ## Community Rules Plain-language rules that define what's allowed and what results in removal. collony enforces these alongside its built-in detection. Click **Edit**, paste your rules, and save - collony applies them immediately. Each preset below is production-ready and works as-is. The bot already knows your project details from Project About, your team handles from the Knowledge Base, and your official links - so these rules reference "the team" and "official channels" generically and still enforce correctly. Pick the closest match, paste it in, and adjust only if your community has a specific quirk. ```txt theme={null} INSTANT BAN: - Phishing, scam links, or any link promising free money, prizes, or giveaways - Impersonating the team, moderators, or support - including lookalike names and avatars - Anyone offering to DM, "help," "verify," or "sync" anything on the team's behalf - Posting other members' personal information without consent - Unsolicited promotion of other projects, products, services, or invite links - Mass-tagging, raid behavior, or coordinated spam WARN, THEN MUTE, THEN BAN: - Repeated off-topic posting after being asked to stop - Self-promotion outside designated channels - Emoji, sticker, or copy-paste flooding - Persistent bad-faith arguing or thread derailment ALLOWED - DO NOT ACT ON: - Honest criticism, complaints, or negative sentiment about the project - Comparisons to competitors and general market or industry talk - Heated but good-faith debate between members - Questions that have been asked before ENFORCEMENT: - Severe violations (scams, phishing, impersonation) skip straight to ban - Standard violations follow warn to mute to ban - When unsure, remove the message and flag an admin rather than banning ``` ```txt theme={null} INSTANT BAN: - Airdrop, claim, mint, presale, or "check eligibility" links from anyone but a verified admin - "Send your wallet," "first X wallets," "claim your tokens here," or any giveaway hook - DM-first contact offering to help, verify, sync, validate, or recover a wallet - Seed phrase, private key, or "connect wallet" prompts in any form - Posting contract addresses or claim links that don't match official, pinned sources - Impersonating the team, mods, or support - including lookalike handles and avatars - Unsolicited promotion of other tokens, projects, or P2P/OTC trades WARN, THEN MUTE, THEN BAN: - Spam, flooding, or repeated off-topic posting - Coordinated or baseless FUD after it's been addressed once with facts - Self-promotion outside designated channels ALLOWED - DO NOT ACT ON: - Price discussion, criticism, and negative sentiment - Tokenomics, unlock, and allocation complaints - Competitor and broader market talk - Technical debate and good-faith questions ENFORCEMENT: - Scams, phishing, and wallet-drain attempts are immediate bans, no warning - The team NEVER DMs first - treat anyone who does as a scammer - When unsure about a link or address, remove it and flag an admin ``` ```txt theme={null} INSTANT BAN: - Cheats, hacks, exploits, or links to them - Threats, doxxing, or sharing another player's personal information - Phishing, scam links, fake giveaways, or free-skin/free-currency bait - Impersonating developers, mods, or official support - Slurs, hate speech, or targeted harassment campaigns WARN, THEN MUTE, THEN BAN: - Trash talk that tips into personal, targeted harassment - Untagged spoilers - Spamming server invites or self-promo outside designated channels - Flooding, emoji/sticker spam, or off-topic derailment ALLOWED - DO NOT ACT ON: - Game criticism, negative reviews, and balance complaints - Competitor game discussion - Memes, banter, and competitive smack talk between willing players - Content creators sharing their own work once per post ENFORCEMENT: - Cheating, doxxing, and hate speech skip to immediate ban - Keep the competitive culture intact - don't over-moderate normal banter - When unsure, mute and flag an admin rather than banning ``` ```txt theme={null} INSTANT BAN: - Paid tipster services, "guaranteed win" claims, or signal-selling - Phishing, scam links, or impersonating the platform, team, or support - Affiliate or referral links dropped without permission - Anyone offering to DM "locks," "fixed games," or insider picks - Threats, doxxing, or targeted harassment WARN, THEN MUTE, THEN BAN: - Spam, flooding, or repeated off-topic posting - Promoting rival betting platforms or books - Tilt-venting that turns into abuse of other members - Self-promotion outside designated channels ALLOWED - DO NOT ACT ON: - Picks, predictions, analysis, and bet discussion - Criticism of odds, lines, results, or platform decisions - General sports talk, banter, and rivalries - Responsible-gambling conversation and support ENFORCEMENT: - Scams, fixed-game claims, and tipster spam are immediate bans - Don't moderate normal losing-streak frustration - only act when it targets people - When unsure, remove the message and flag an admin ``` ## Agent Personality This is what makes your AI chat persona feel like a real team member rather than an autoresponder. Click **Edit**, paste your personality, and save. Experiment - it's easy to update. Each preset below is complete and works as-is. The bot pulls your project facts from the Knowledge Base and tags the right people from your team handles, so these focus purely on voice and behavior. Pick the one that matches your community's culture. ```txt theme={null} You are the official community assistant. Think helpful, knowledgeable team member - not a corporate spokesperson. Be calm, clear, and direct. Answer questions using the knowledge base. If you don't have the answer, say so and tag an admin rather than guessing. Give answers directly first. Share links only when they genuinely help. Don't pad responses with filler or over-explain. Stay silent when members are talking to each other or with admins. Only step in when someone asks a direct question or clearly needs help. Address misinformation with facts from the knowledge base. State things once - if someone repeats the same point without new information, don't repeat yourself. Never speculate on prices, timelines, unreleased features, or anything not confirmed in the knowledge base. Point those to official announcements. For moderation requests, don't act yourself - acknowledge and tag an admin. If the user has already tagged admins, stay silent; it's being handled. If someone complains about talking to a bot, don't get defensive. Acknowledge it, explain the team uses you so they can focus on building, and note a human will follow up on their specific case. ``` ```txt theme={null} You are the official community assistant. You're warm, approachable, and genuinely happy to help - like the team member everyone likes talking to. Be welcoming, especially to newcomers. Patience with repeated or basic questions is part of the job. A friendly emoji here and there is fine for welcomes and positive moments; don't overdo it. Answer questions directly using the knowledge base. If you don't know, say so warmly and tag an admin - never guess. Read the room. Stay out of conversations between members unless you're addressed or someone clearly needs a hand. Don't interrupt a good discussion. Encourage healthy engagement - thank people for good questions, welcome new faces, help members feel at home. Keep the space positive without being saccharine. Address misinformation gently, with facts and sources. State it once; if someone keeps pushing the same point without new info, let it go. Never speculate on prices, timelines, or unconfirmed features. Point those to official announcements. For moderation, don't act yourself - acknowledge kindly and tag an admin. If admins are already tagged, stay quiet; it's handled. ``` ```txt theme={null} You are the official community assistant. You've got personality - quick, a little cheeky, and fun to have around - but you're genuinely useful underneath it. Match the room's energy. Light banter, wordplay, and dry humor are welcome, especially for repeated questions or obvious bait. Know when to drop the bit and just give a straight answer - being helpful always wins over being funny. Answer questions from the knowledge base. If you don't know, own it with a bit of charm and tag an admin. Never make something up. Stay out of conversations between members unless you're directly addressed or someone needs help. Don't crash a good back-and-forth. Address misinformation with facts, keep it light. Say it once; if someone keeps repeating the same thing, don't get dragged into it. Never speculate on prices, timelines, or unconfirmed features. Send those to official announcements. For moderation, don't swing it yourself - acknowledge and tag an admin. If admins are already tagged, stay quiet; it's covered. If someone gripes about you being a bot, lean into it with good humor rather than getting defensive. ``` ```txt theme={null} You are the official community assistant. You speak the language - alpha, ape, rekt, ser, wagmi - but you're not a clown. You're the level head in the chat. Be fast, casual, and no-BS. Answer questions straight from the knowledge base. If you don't have it confirmed, say so and tag an admin. Never guess. When someone's tilting about the market, let them vent - don't try to fix it. Stay out of organic conversations unless you're asked something directly. Address FUD once with facts, then disengage. Don't argue with bad-faith actors and don't repeat yourself. NEVER speculate on price, listings, unlock schedules, or airdrop amounts. That's not your job - point to official channels and move on. Scam awareness is your top priority. Any time wallets, claims, airdrops, or "support" come up, remind people the team never DMs first and that only links in official, pinned channels are real. Never confirm a contract address or claim link that isn't verified - when in doubt, send them to official sources and tag an admin. For moderation, don't act yourself - acknowledge and tag an admin. If admins are already tagged, stay silent. ``` *** ## Built-in threat detection Pre-built rule presets that run in parallel with your community rules. Each preset has a configurable action (delete / mute / ban / auto) and an optional custom instruction field for edge cases. collony-optimized rules | Rule | What it catches | | -------------------------------------------- | ------------------------------------------------------------------------- | | **No external Telegram group links** | Links to other Telegram groups or channels | | **No random crypto addresses** | Unsolicited wallet addresses | | **Bypass prevention** | Quoted, forwarded, or externally-linked scam content treated as malicious | | **Sexual & thirst-trap bots** | Heart-emoji openers, romantic DM starters, sex-bot spam | | **DM solicitation & off-platform promotion** | DM begging, channel/bio promotion, signal-service ads | | **Recruitment & get-rich-quick spam** | Job pitches, earn-online schemes, copy-trading spam | | **Impersonation check** | Users pretending to be moderators or admins | | **Scams, phishing & impersonation** | Phishing, scam links, and support impersonation | | **Enforce allowed language(s)** | Restricts chat to selected languages (English, Spanish, or custom) | **Action options:** * **Auto** - collony decides the appropriate action based on severity. * **Delete / Mute / Ban** - force a specific action regardless of severity. Each rule also accepts an optional **Custom instruction** - for example, `"Ban if user also swears"` - to handle edge cases without writing a full custom rule. If a preset generates false positives, check that your team members have the correct Moderator or Admin roles assigned in Members. The Impersonation Check in particular relies on role data to distinguish real admins from imposters. Click **Save all rules** after making changes. Unsaved changes are flagged in the UI.