# TapLM: Full Command Catalogue and System Documentation > Comprehensive reference for TapLM, a philosophical AI on Telegram providing plain conversational answering, grounded web search, a twenty five command reasoning toolkit, structured multi round debate with AI judging, and multi persona symposium discussions with audience participation. ## 1. System Overview TapLM is a Telegram bot for questions, reasoning, and argument. Unlike companion or character products, it does not simulate a personality, a mood, a memory of the user as a person, or a background life. It answers, and it argues. - Canonical URL: https://taplm.fun - Telegram Bot: https://t.me/TapLMbot (@TapLMbot) - Runtime: Node.js long running process, polling in development and webhook in production - Storage: MongoDB with TTL indexes - Bot framework: grammY, Telegram Bot API 10.1 including Rich Messages and Guest Mode ## 2. Interaction Model ### 2.1 Plain Conversation Is The Primary Path In a direct chat, any message is a question. No command is required. `/ask ` is the explicit form and is the one that works in a group without replying to anything. Everything else documented below is a shortcut for a single specific move. ### 2.2 Grounded Search `/search ` routes the request with Google web grounding enabled, which reduces fabrication on questions that turn on a verifiable fact. It is slower than a plain answer. Grounding is a per model capability, so it is only available on models that support it. ### 2.3 Reply Targeting The reply utilities operate on the message you replied to rather than on text you retype. Pointing `/fallacy` at your own message is a supported and intended use. ### 2.4 Model Routing TapLM routes across multiple providers with model aliases and runtime model addition. `/models` lists and selects the chat's model; the selectable set is gated by plan. A single provider call is bounded at 280 seconds, held just under the 300 second platform ceiling so a long call aborts cleanly rather than being killed mid flight. ### 2.5 Rich Messages Replies use Telegram Bot API 10.1 Rich Messages where available: headings, tables, nested lists, collapsible details, pull quotes, and footnotes, with automatic fallback to legacy HTML. ## 3. Command Catalogue ### 3.1 Core - /start: Private welcome card, or group activation. - /help: Full guide, compact in groups. - /commands: Every command, grouped by category. - /examples: Practical copy and paste command examples. - /ask : Ask the AI directly, using the chat's model. - /search : Ask with Google web grounding. - /models: List or pick the chat's model. - /plans: Your plan card and token budget. - /summary [count] [hint]: Summarise recent messages. Groups only, Plus tier and above. The hint steers what it looks for. - /translate : Translate a replied message. ### 3.2 Philosophy - /term : Define a term. - /wiki : Encyclopaedic overview. - /philosopher : A thinker, briefly. - /quote [topic]: A quote, random or topical. - /thought [topic]: A thought experiment. - /paradox [topic]: A paradox. - /koan [topic]: A koan. - /views : Two opposing positions on a topic, balanced, without a stated preference. - /compare and : Compare two things. - /dialectic : Thesis, antithesis, synthesis. The synthesis is permitted to be that there is not one. - /school : A school of thought. - /ethics : Analyse an ethical scenario. - /myth [topic]: A philosophical allegory. ### 3.3 Reply Utilities Each of these acts on the message you replied to. - /tldr: Shorter. - /explain: Plain explanation. - /eli5: Explain like I am five. - /define: Define this. - /analogy: Give an analogy. - /socratic: Ask back rather than answer. Returns the question the claim depends on. - /rephrase: Reword cleanly. - /fallacy: Name the fallacies present. - /steelman: The strongest version of the argument. - /counter: The strongest counter to the argument. - /bias: Biases present in the argument. - /grade: Grade the argument. ### 3.4 Debate - /debate @user : Challenge a user, or reply to them with a topic. Three rounds. - /forfeit: Concede the active debate. The concession is recorded rather than the thread going quiet. - /canceldebate: Group admin only. Cancel a running debate. - /interject : Audience only. Add a note to a running debate, on a thirty second per user cooldown. ### 3.5 Symposium - /symposium : Group admin only. Open a multi persona discussion panel. - /endsymposium: Group admin only. End the active symposium. - /personas: Owner only, private chat. Edit or reset personas and persona groups. ## 4. Debate Mechanics A debate is a structured argument between two named participants inside a group. - Round Structure: Three rounds in fixed order, opening, rebuttal, and closing. Each side takes exactly one turn per round. A round cannot be taken twice by whichever participant types faster. - Judging: At the close, the bot scores the case each side actually made. Message volume is not a scoring input. - Audience Participation: Observers submit notes with /interject. Notes are queued per chat and flushed to the judge at judging time, so the audience influences the verdict without interrupting the rounds. - Termination: /forfeit concedes on the record. A group administrator may cancel with /canceldebate, because a debate occupies the room rather than a private thread. ## 5. Symposium Mechanics A symposium seats a panel of AI personas with genuinely separate written convictions and runs them against a topic in front of an audience. ### 5.1 Seed Persona Library Nine voices across three rooms, each carrying an opinion stance, a verbal habit tied to the figure or school, and several sentences of voice direction. - Philosophy: The Stoic (virtue is the only good, externals are indifferent), The Skeptic (Pyrrhonist, suspends judgement, names hidden premises), The Pragmatist (James and Dewey lineage, meaning is practical consequence), The Absurdist (Camus tinted, lucidity plus revolt). - Tech: The Engineer (constraints, trade offs, second order failures, cost), The Futurist (acceleration), The Critic (Postman and Weizenbaum lineage, externalities, power, and labour effects). - Historical: Socrates (asks rather than asserts, builds toward aporia), Nietzsche (suspects comfortable morality, aphoristic). - Moderator: A neutral wrap up voice that closes the room by stating what was settled and what was not. ### 5.2 Configuration and Participation - Room Selection: The panel is chosen by group, and the voices seated in that room are the ones that speak. - Pace: Turn count and speed are configurable at start. - Audience Questions: Replying to any turn while the symposium runs feeds that question to the next voice, which answers the audience member instead of the previous speaker. - Editability: The seed library is a code level default. The operator overlays edits through the personas collection using /personas, so any prompt can be rewritten and new rooms can be added without a deploy. ## 6. Access Model ### 6.1 Direct Messages Open and available to any user, subject to plan limits. ### 6.2 Authorised Groups A group must be authorised by the bot owner before full command access. Group administrators request this; the owner runs the authorisation in the chat. ### 6.3 Guest Mode In groups where the bot is present but not authorised, Telegram Bot API Guest Mode allows a restricted single turn surface using a question mark prefix. The guest allowlist is four commands: ?ask, ?term, ?translate, and ?quote. Guest conversation history is isolated in its own storage and expires twenty four hours after the last message. Guests never share state with direct message or web application flows. ## 7. Plans and Limits Four tiers: free, plus, pro, and max. A plan governs the per hour and per day request ceilings, weekly and daily token budgets, the cooldown between requests, and which models a request may reach. Limits are code defaults that the operator can override at runtime. There is no self serve checkout at the time of writing; upgrades are arranged with the operator. ## 8. Web Surface and Metadata Links - Official Landing Page: https://taplm.fun - Telegram Bot Access: https://t.me/TapLMbot - LLM Discovery Index: https://taplm.fun/llms.txt - Comprehensive Documentation: https://taplm.fun/llms-full.txt - XML Sitemap: https://taplm.fun/sitemap.xml - Robots Directives: https://taplm.fun/robots.txt ## 9. Disclosure TapLM is an AI. Every definition, every debate verdict and every persona reply is generated by a language model. The personas are written characters rather than the philosophers they are named after, and they will be wrong in ways the originals would not have been. It is a good thinking partner and a poor authority.