When Your Plant Talks Back: Conversational AI with InfluxDB 3

By Suyash Joshi Developer
Mar 10, 2026

No one wants to stare at a plant and guess if it needs water. It’s much easier if the plant can say, “I’m thirsty.” A few years ago, we built Plant Buddy using InfluxDB Cloud 2.0. The linked article is still a great guide for cloud-first IoT prototyping as it shows how quickly you can connect devices, store time series data, and build dashboards in the cloud with the previous version of InfluxDB.

But this time, the goal was different. Instead of sending soil moisture data to the cloud, the entire system runs locally using the latest InfluxDB 3 Core, similar to a modern industrial setup powered by LLM for a natural conversational interaction.

The architecture: the “factory” at home

In real factories, raw PLC data is captured at the edge, often using MQTT and a local database. That same architecture now powers PlantBuddy v3 with the following setup.

• Edge Device (ESP32 / Arduino): Works like a small PLC. It reads soil moisture and publishes the plant’s state to the network without knowing anything about the database.
• Soil Moisture Sensor (Analog): Outputs an analog signal based on soil moisture. The microcontroller converts it to digital using its built-in ADC.
• Message Bus (Mosquitto MQTT): Handles publish/subscribe communication. The Arduino publishes data to the broker (running locally), and Telegraf subscribes to forward data to the database.
• Database (InfluxDB 3 Core): Runs locally in Docker as a high-performance time series database storing all sensor readings.
• User Interface (Claude + MCP): Enables natural language queries. Instead of writing SQL, questions about plant health can be asked conversationally.

1. Collecting & Sending Data from the Edge

To make this scalable, I treat the sensor data like an industrial payload. It’s not just a number; it’s a structured JSON object containing the ID, raw metrics, and a pre-calculated status flag.

The Arduino Payload

{ 
"id": "pothos_01",    // Device identifier (like a PLC tag) 
"raw": 715,  		// Raw ADC value (0-1023) 
"pct": 19,  		// Calculated moisture percentage 
"stat": "DRY_ALERT"   // Pre-computed status 
}

Why compute status at the edge? In factories, PLCs make local decisions (e.g., stop motor, trigger alarm). Here, the Arduino determines “DRY_ALERT” so the database can trigger alerts without recalculating thresholds.

2. The Ingest Pipeline

Below are two approaches to send plant data to InfluxDB. In this project, I went with MQTT and Telegraf, which are more standard for an industrial setup.

Telegraf acts as the gateway, subscribing to the MQTT broker and translating the JSON into InfluxDB Line Protocol. This configuration is identical to what you’d see in a manufacturing plant monitoring vibration sensors.

# telegraf.conf - Complete Working Example
[agent]
  interval = "10s"
  flush_interval = "10s"

[[inputs.mqtt_consumer]]
  servers = ["tcp://127.0.0.1:1883"]
  topics = ["home/livingroom/plant/moisture"]
  data_format = "json"

  # Tags become indexed dimensions (fast filtering)
  tag_keys = ["id", "stat"]

  # Fields become measured values
  json_string_fields = ["raw", "pct"]

[[outputs.influxdb_v2]]
  urls = ["http://127.0.0.1:8181"]
  token = "$INFLUX_TOKEN"
  organization = "my-org"
  bucket = "plant_data"

Note: If Telegraf runs in Docker, use host.docker.internal:8181 to reach the database.

3. Time Series Database: InfluxDB 3 (Docker Container)

InfluxDB 3 Core runs locally in Docker as the time series database. It stores soil moisture readings and enables real-time analytics, all without depending on external cloud connectivity.

# Create persistent storage 
mkdir -p ~/influxdb3-data

# Run InfluxDB 3 Core with proper configuration
docker run --rm -p 8181:8181 \
  -v $PWD/data:/var/lib/influxdb3/data \
  -v $PWD/plugins:/var/lib/influxdb3/plugins \
  influxdb:3-core influxdb3 serve \
    --node-id=my-node-0 \
    --object-store=file \
    --data-dir=/var/lib/influxdb3/data \
    --plugin-dir=/var/lib/influxdb3/plugins

4. The “AI” Interface: InfluxDB MCP & Claude

Instead of writing SQL queries or building dashboards, the system connects an LLM to InfluxDB through the Model Context Protocol (MCP). I’ve written another blog post on how to connect InfluxDB 3 to MCP, which you can find here.

Now the question isn’t: “What’s the SQL query for average soil moisture over the last 24 hours?”

It becomes: “Has the plant been dry today?”

The LLM generates the correct SQL under the hood. If needed, the generated query can be inspected. This makes time series analytics accessible through conversation.

claude_desktop_config.json

{
  "mcpServers": {
    "influxdb": {
      "command": "docker",
      "args": [
        "run",
        "--rm",
        "--interactive",
        "--add-host=host.docker.internal:host-gateway",
        "--env",
        "INFLUX_DB_PRODUCT_TYPE",
        "--env",
        "INFLUX_DB_INSTANCE_URL",
        "--env",
        "INFLUX_DB_TOKEN",
        "influxdata/influxdb3-mcp-server"
      ],
      "env": {
        "INFLUX_DB_PRODUCT_TYPE": "core",
        "INFLUX_DB_INSTANCE_URL": "http://host.docker.internal:8181",
        "INFLUX_DB_TOKEN": "YOUR_RESOURCE_TOKEN"
      }
    }
  }
}

The Result:

What’s next

In the next post, we will upgrade this Plant Buddy project to do more than passively monitor. New features will include:

• A water pump, motor, and small display.
• Automatic watering when the plant enters DRY_ALERT.
• An extended system with light and temperature sensors to determine how placement of the potted plant affects its health, especially during trips when no one is home.

Try to build one yourself with InfluxDB 3! We would love to hear your questions/comments in our community forum, Slack, or Discord.