Multi-Agent Dynatrace Observability Showcase

LLM-driven multi-agent system for Telemetry, Problems, Security, and DevOps. Orchestrated with LangGraph. Data via the Dynatrace MCP server. UI with Gradio, optional CLI.

Architecture

The system is organized into three layers: a top-level supervisor and domain-specific teams.

1. Teams Supervisor

• Routes user requests to one or more domain teams: Telemetry, Problems, Security, DevOps.
• Ensures each team is called at most once.
• Always ends with FINISH after at least one team responded.

2. Domain Teams

Each domain team has its own supervisor and two workers: a Fetcher and an Analyst.
Workflow: Fetcher → Analyst → FINISH

Telemetry Team

• Fetcher: retrieves logs, metrics, spans via Dynatrace MCP tools.
• Analyst: analyzes anomalies, correlates signals, suggests mitigation.

Problems Team

• Fetcher: fetches active problems (list_problems).
• Analyst: identifies root causes, impact, prioritizes issues.

Security Team

• Fetcher: fetches vulnerabilities (list_vulnerabilities).
• Analyst: ranks risks, groups CVEs, highlights exposure, proposes mitigations.

DevOps Team

• Fetcher: retrieves deployment events, SLO/SLI data, error budgets.
• Analyst: evaluates health gates, canary rollbacks, error budget status, suggests remediation.

3. Tools

• MCP Tools: Dynatrace MCP
• Retrievers
• FAISS index from dynatrace LLM rules

4. Guardrails

• Active on all user input.
• Blocks toxic language, profanity, banned terms.
• On violation, returns safe response instead of executing requests.

5. Execution Flow

1. User input → Teams Supervisor decides which team to call.
2. Domain Supervisor runs: Fetcher → Analyst.
3. Analyst output → back to Teams Supervisor.
4. Teams Supervisor may call another team or route to FINISH.
5. Final combined response returned to user.

Requirements

• uv package manager (pipx install uv or pip install uv)
• OpenAI API key
• Dynatrace environment URL and platform token

Quickstart

1. Initialize project

   uv init
   

2. Install Guardrails

   scripts\install_guardrails.bat
   

3. Create .env in repo root and fill values Create a file named .env at the repository root with:

   env

   OpenAI

   OPENAI_API_KEY=sk-...
   

   Dynatrace

   DT_ENVIRONMENT=https://<env>.apps.dynatrace.com
   DT_PLATFORM_TOKEN=dtp_...
   DT_ACCOUNT_PW=
   

   Tracing / Observability

   TRACELOOP_BASE_URL=https://<tenant>.live.dynatrace.com/api/v2/otlp
   TRACELOOP_HEADERS=Authorization=Api-Token <YOUR_TOKEN>
   OTEL_EXPORTER_OTLP_METRICS_TEMPORALITY_PREFERENCE=delta
   

   LangSmith

   LANGSMITH_TRACING=
   LANGSMITH_ENDPOINT=
   LANGSMITH_API_KEY=
   LANGSMITH_PROJECT=
   

   Guardrails

   GUARDRAILS_API_KEY=
   

4. Start Gradio showcase

   scripts\start_gradio_showcase.bat
   

   Open the printed local URL, default http://127.0.0.1:7860.

Optional: start CLI

scripts\start_cli_showcase.bat

Scripts

• scripts/install_guardrails.bat — install Guardrails and hub deps
• scripts/start_gradio_showcase.bat — launch the Gradio UI
• scripts/start_cli_showcase.bat — launch the CLI
• scripts/start_dynatrace_mcp.bat — helper to run the Dynatrace MCP server