Instrumenting AI/LLM Chatbots with Observability using TrueWatch, Dify, and Ollama

(Hands-on demo for TrueWatch users & partners)

This guide helps you instrument your AI/LLM chatbot (built on the Dify platform with Ollama LLM) with observability using TrueWatch and DataKit.

Prerequisites

• Machine (e.g., Ubuntu 24.04) with at least 10 GB RAM and 2 vCPUs.
• Create an account at TrueWatch.
• Create a workspace in TrueWatch and obtain your workspace token:
  • Log in to TrueWatch, select your workspace, and copy your workspace token (ENV_DATAWAY).

Step 1: Install Docker and Docker Compose

sudo apt update
sudo apt install docker.io docker-compose -y
sudo systemctl start docker
sudo systemctl enable docker

Step 2: Clone and Set up Dify Platform

git clone https://github.com/langgenius/dify.git
cd dify/docker

Step 3: Integrate TrueWatch Observability and DataKit

Clone the TrueWatch Git repository containing necessary configuration files:

git clone https://github.com/TrueWatchTech/llm-observability-demo-setup-guide.git

Copy and overwrite the necessary configuration files from the cloned TrueWatch repo to the Dify Docker folder:

cp -R llm-observability-demo-setup-guide/resources/datakit-config ./
cp llm-observability-demo-setup-guide/resources/docker-compose.yaml ./
cp llm-observability-demo-setup-guide/resources/Dockerfile.proxy ./
cp llm-observability-demo-setup-guide/resources/app.py ./

Ensure your docker-compose.yaml includes your correct workspace token under the datakit service (line 564):

environment:
  ENV_DATAWAY: "https://<your_truewatch_region>-openway.truewatch.com?token=<your_workspace_token>"

Step 4: Deploy and Run Containers

docker compose up -d

Step 5: Access and Configure Dify

1. Open your browser and navigate to: http://localhost.
2. Create a workspace.
3. Set up your LLM (Ollama):
   • Navigate to Models → Add Model.
   • Configure your Ollama model URL (e.g., http://ollama:11434). Use default settings for the rest of the fields.

Step 6: Deploy the Proxy (app.py)

app.py acts as a proxy and is invoked via the Dockerfile.proxy file. Ensure the app.py and Dockerfile.proxy are correctly placed in your docker‑compose setup as shown in Step 3.

Note: If you make changes to app.py, you do not have to redeploy all services; just redeploy the proxy service as shown in the Troubleshooting section below.

Step 7: Preview and Interact with the Bot

• In the Dify workspace, select your chatbot and click Preview.
• Interact with the chatbot to generate traces and data for TrueWatch observability.

Step 8: View Observability Data in TrueWatch

1. Log in to your TrueWatch account.
2. Navigate to APM → Traces.
3. Filter services by ollama-proxy to view detailed traces and spans, including token usage and costs.

Step 9: Explore APM Traces (placeholders for screenshots)

Follow the numbered callouts in the screenshot below to understand each section of a trace:

Step 10: Use the Pre‑built LLM Observability Dashboard

TrueWatch ships with a ready‑made dashboard that aggregates key LLM usage metrics (token counts, usage over time, etc.)

1. Go to Dashboards → Create → System Views → OpenLIT Monitor View.
2. Furnish the fields to create your OpenLIT Monitor View dashboard.
3. Select the time range that matches your chatbot’s activity.
4. Hover over any panel for tooltips and click a chart to drill into traces behind a spike.

Import Custom Dashboard

You may also import a pre-built custom dashboard that includes more metrics such as success/error rates, call response times, etc.

1. Go to Dashboards → Import Dashboard → Open llm-observability-demo-setup-guide/resources/LLM Observability (OpenLIT) Dashboard.json.
2. Import Pipeline Script via Management → Pipelines → Import llm-observability-demo-setup-guide/resources/Pipelines Template.json.

Troubleshooting and Validation

Redeploy Only the Ollama Proxy (after app.py updates)

docker compose up -d --build ollama

Check Connectivity to Ollama Service

docker exec -it docker-api-1 bash
curl -v http://ollama:11434/v1/models
curl -v http://ollama:11434/v1/chat/completions   -H "Content-Type: application/json"   -d '{"model":"llama3","messages":[{"role":"user","content":"ping"}]}'

Check Ollama Logs for Errors

docker compose logs --tail=50 ollama

Ensure no Host‑side Ollama or DataKit Services Are Running

Host‑installed versions can occupy the same ports (11434 for Ollama, 9529/4317/4318 for DataKit) and block the containers.

# --- Ollama ------------------------------------------------------
sudo lsof -i :11434        # native process
sudo systemctl stop ollama
sudo systemctl disable ollama
sudo lsof -i :11434        # should return nothing
 
# --- DataKit -----------------------------------------------------
sudo lsof -i :9529         # DataKit UI
sudo lsof -i :4317         # OTLP gRPC
sudo lsof -i :4318         # OTLP HTTP
sudo systemctl stop datakit
sudo systemctl disable datakit
sudo lsof -i :9529 :4317 :4318

Managing Docker Compose

Next Steps

• Customize observability dashboards in TrueWatch.
• Integrate additional instrumentation such as logs and metrics.

References

• TrueWatch Documentation
• Dify Platform Documentation
• Ollama Documentation