IchiBot

Overview

IchiBot is where the chatbot obsession started: a Telegram bot built in 2020 on AIML, the rule-based markup language that powered a whole generation of pre-LLM conversational agents. No transformers, no embeddings. Just pattern-matching categories, carefully written by hand, deployed as @IcHiGo_bot on a self-hosted server.

It held surprisingly natural conversations across everyday topics, and building it meant learning the hard version of the craft first: when your bot can only say what you explicitly taught it, you learn very quickly how people actually phrase things.

Architecture

flowchart LR
    USER["Telegram user"]
    TG["Telegram Bot API<br/>@IcHiGo_bot"]
    subgraph HOST["Self-hosted server"]
        ROUTER["Message handler"]
        AIML["AIML Engine<br/>pattern → template matching"]
        KB["AIML Categories<br/>hand-authored topic files"]
    end
    USER --> TG --> ROUTER --> AIML
    AIML --> KB
    KB --> AIML
    AIML -->|response| TG --> USER

Engineering Decisions

• AIML over scripting from scratch. Learning the markup properly (patterns, templates, srai redirects for synonym folding) gave one rule file the reach of dozens of hardcoded branches.
• Self-hosted deployment. The bot ran on my own server rather than a managed platform, which meant owning uptime, process management, and the Telegram webhook lifecycle end to end.

Highlights

• Deployed and public on Telegram as @IcHiGo_bot
• Hand-authored AIML knowledge base covering diverse everyday topics
• The origin point of a line that runs straight through to today’s agent work: same problem, five generations of better tools