spring-boot-dentist-agent

Atmosphere @Agent — Dr. Molar (Dental Emergency Assistant)

A multi-channel dental emergency assistant built with @Agent. One agent class — the same @Command slash commands and AI tools work on Web, Slack, Telegram, Discord, WhatsApp, and Messenger simultaneously. A patient types /firstaid in Slack and gets the same instant response as in the browser.

What It Does

Dr. Molar helps patients with broken, chipped, or cracked teeth. It can assess emergency severity, recommend pain management, and provide first-aid instructions. The agent works on every configured channel at once — no per-channel code needed.

Slash Commands (work on every channel)

Command	Description
/firstaid	Step-by-step first-aid for a broken tooth
/urgency	Help determine how urgently you need a dentist
/pain	Pain management tips (OTC options, home remedies)
/help	List all commands (auto-generated)

These commands are defined once with @Command in DentistAgent.java. When atmosphere-channels is on the classpath, they automatically appear on every channel — Slack, Telegram, Discord, WhatsApp, Messenger. Natural language messages on those same channels fall through to the LLM via @Prompt.

AI Tools

Tool	Parameters	Description
assess_emergency	injury_type, pain_level, bleeding	Classify severity: EMERGENCY, URGENT, SAME-DAY, SOON, ROUTINE
pain_relief	pain_level, allergies	Recommend pain management accounting for allergies

Running

# Demo mode (no API key needed)
atmosphere run spring-boot-dentist-agent

# Or from the repository root
./mvnw spring-boot:run -pl samples/spring-boot-dentist-agent

# With a real LLM
LLM_API_KEY=your-gemini-key ./mvnw spring-boot:run -pl samples/spring-boot-dentist-agent

Open http://localhost:8080/atmosphere/console/ in your browser.

With Channels

# Slack
SLACK_BOT_TOKEN=xoxb-... SLACK_SIGNING_SECRET=... \
  ./mvnw spring-boot:run -pl samples/spring-boot-dentist-agent

# Telegram
TELEGRAM_BOT_TOKEN=... TELEGRAM_WEBHOOK_SECRET=... \
  ./mvnw spring-boot:run -pl samples/spring-boot-dentist-agent

Try These

• /firstaid — instant 7-step first-aid guide (no LLM call)
• /urgency — triage checklist: ER, same-day, 1-2 days, or routine
• /pain — OTC meds, home remedies, and what to avoid
• I chipped my front tooth and it hurts (6/10), no bleeding — LLM invokes assess_emergency
• I'm allergic to ibuprofen, pain level 7 — LLM invokes pain_relief with allergy awareness

Key Code

File	Purpose
DentistAgent.java	@Agent with @Command and @AiTool methods
prompts/dentist-skill.md	System prompt with personality, skills, tools, channels, guardrails
LlmConfig.java	Configures LLM backend settings (model, API key, base URL)
ConsoleEndpointAlias.java	Aliases the agent handler to /atmosphere/ai-chat for the console
DemoResponseProducer.java	Fallback responses when no LLM API key is set

Architecture

Web browser ──WebSocket──┐
Slack ────────webhook────┤
Telegram ─────webhook────┤   @Agent(name = "dentist")
Discord ──────gateway────┤            │
WhatsApp ─────webhook────┤       CommandRouter
Messenger ────webhook────┘        ┌───┴───┐
                            "/" prefix?   natural language
                                  │              │
                             @Command       @Prompt ──> LLM + @AiTool
                             (instant)                  (assess_emergency,
                                                         pain_relief)

The same @Command methods and @Prompt handler serve every channel. /firstaid in Slack returns the exact same first-aid steps as in the browser. The atmosphere-channels module handles message format differences (max length, reply threading, markdown support) transparently.

Skill File

The prompts/dentist-skill.md defines Dr. Molar's persona and capabilities:

# Dr. Molar — Emergency Dental Assistant
You are Dr. Molar, a friendly and knowledgeable dental emergency assistant.

## Skills          <-- extracted for A2A Agent Card
## Tools           <-- cross-referenced with @AiTool methods
## Channels        <-- validated against classpath (slack, telegram, web)
## Guardrails      <-- LLM self-enforces (always state you're an AI, never diagnose)

The entire file becomes the system prompt. Sections are also parsed for protocol metadata (A2A, MCP).

Environment Variables

Variable	Default	Description
LLM_API_KEY	(none)	API key for Gemini, OpenAI, or compatible. Omit for demo mode.
LLM_MODEL	gemini-2.5-flash	Model name
SLACK_BOT_TOKEN	(none)	Slack bot token (xoxb-...)
SLACK_SIGNING_SECRET	(none)	Slack app signing secret
TELEGRAM_BOT_TOKEN	(none)	Telegram bot token from @BotFather
TELEGRAM_WEBHOOK_SECRET	(none)	Random string for webhook verification