Assist documentation.

Connect your Dynatrace environment

Getting started

1. Provision an API token
In your Dynatrace environment, create an API token for Assist. Assist reads operational data — metrics, problems, logs, traces and topology — to answer questions and run investigations. It does not write to your Dynatrace environment.

2. Point Assist at the environment
Supply your Dynatrace environment URL and the token from step 1. Assist is built for Dynatrace Managed and fully on-premises estates, including air-gapped deployments.

3. Connect your directory
Assist maps users and roles from your existing directory — Entra ID for the Microsoft Teams surface, LDAP for the on-premises web UI. See Surfaces & access control.

4. Make a model available
Assist runs its LLM locally on your own infrastructure, so no observability data or prompts are sent to external AI services. Which models are approved for the environment is your decision — see Security & data handling.

First login

What you'll see

Assist opens on a conversation, not a dashboard. You type an operational question in plain language; Assist classifies the intent, routes it to the right pipeline — live query, root-cause investigation, or knowledge-base chat — and answers with evidence drawn from your environment.

In the web UI, answers involving metrics come back as interactive charts alongside the explanation. In Microsoft Teams, the same capabilities are available through the bot and its slash commands.

Asking questions

Core concept

You do not need to start from a dashboard, a DQL query, or a predefined navigation path. Assist takes the operational question as the starting point and plans the Dynatrace API calls needed to answer it.

Questions that work well:

• “Why is this service slowing down?”
• “What changed before the incident started?”
• “Which dependencies are most likely contributing to the problem?”
• “Summarize the impact for the service owner.”
• “Show me open problems.” / “Error rate for the last hour.”

Follow-up questions narrow the problem or test a new hypothesis without restating context. You can also ask Assist to re-present the same finding for a different audience — the technical detail an SRE needs, or the business-level explanation an executive needs.

No DQL required. DQL still matters. Assist reduces the need for every user to know DQL before they can benefit from Dynatrace data; it does not remove the platform's depth for the people who do.

Agentic investigation

Core concept

An investigation is rarely one question. Given an instruction such as “Investigate the degradation affecting our payment service,” Assist can coordinate the steps as an agentic workflow around your intent:

• Define the scope — affected service, timeframe, environment and symptoms.
• Inspect service health — performance, errors, dependencies and events.
• Look for change — deployments, configuration changes and anomalies around the start of the issue.
• Correlate evidence — signals that otherwise live across different Dynatrace views.
• Assess impact — which services, users or business transactions appear affected.
• Summarize findings — the relevant evidence, and the questions still open.
• Recommend next steps — further investigation, or operational action.

The root-cause pipeline enriches a Dynatrace problem with topology, deployments, logs and traces in parallel, recalls the most similar past incidents, and produces a report with ranked, evidence-cited hypotheses rather than a single unexplained answer.

Assist is built to show its work: what question it understood, what Dynatrace information it examined, what it found, what it is inferring, where uncertainty remains, and what it recommends doing next.

Briefings & reports

Feature

Assist generates operational summaries over any time window — an incident write-up, a service summary, or a morning status briefing. Briefings export to PDF and can be delivered on a schedule by email, so a recurring ops review is produced rather than written by hand.

Knowledge base

Feature

Upload your runbooks, SOPs and postmortems, and Assist grounds answers in your own documentation alongside live Dynatrace data. Completed investigations feed back into that knowledge, so each root-cause analysis builds an incident memory the next one can draw on.

Surfaces & access control

Reference

Assist runs in three modes, with the same capabilities behind each surface:

• Microsoft Teams — native bot with slash commands, identity from Entra ID.
• On-premises web UI — identity from LDAP.
• Hybrid — both at once.

Access uses three roles — viewer, SRE and admin — assigned from your directory groups. Sensitive capabilities such as root-cause analysis, model switching and administration are gated per role. Every action is rate-limited and audit-logged.

Deployment

Reference

• Docker Compose — for a fast on-premises start.
• Helm chart — for Kubernetes.
• Secrets — held in HashiCorp Vault.
• Observability — health, metrics and OpenTelemetry endpoints are built in, so you can monitor Assist with the platform you already run.

Deployment models — on-premises, sovereign, private and hybrid, or with approved external AI services where policy permits — are agreed as part of onboarding. Get in touch to scope yours.

Security & data handling

Reference

• Nothing leaves your environment — the LLM runs locally on your infrastructure; no observability data or prompts are sent to external AI services. Air-gapped environments are fully supported.
• Your own persistence — all data is stored in your own PostgreSQL.
• Read-only Dynatrace access — Assist reads from your Dynatrace environment to investigate; it does not write to it.
• Gated LLM calls — every call passes through a single audited gateway, with injection scanning and PII redaction.
• Audit logging — interactions and actions are logged to both file and database.

Assist is designed to support a progression from insight toward action — explain, recommend, prepare, request approval, trigger a governed workflow — without assuming that every action should become autonomous. For sensitive operations, human approval remains part of the workflow, and what is permitted is yours to configure.

Integrations

Reference

FAQ

Does any of our data leave our environment?
No. Assist is designed for on-premises deployment: the LLM runs locally on your infrastructure, all persistence is in your own PostgreSQL, and Dynatrace access is read-only. Air-gapped environments are fully supported.

Does Assist work in Teams, web, or both?
Three modes: Microsoft Teams (Entra ID identity), an on-premises web UI (LDAP identity), or hybrid running both at once. Same capabilities behind every surface.

How is access controlled?
Three roles — viewer, SRE and admin — assigned from your directory groups. Sensitive capabilities like RCA, model switching and administration are gated per role, and every action is rate-limited and audit-logged.

How does Assist deploy?
Docker Compose for a fast on-premises start, with a Helm chart for Kubernetes. Secrets live in HashiCorp Vault, and health, metrics and OpenTelemetry endpoints are built in for your own monitoring.

Do users need to know DQL?
No. Assist takes the operational question as its starting point and plans the queries underneath. Experienced users keep the full depth and precision of DQL and the Dynatrace platform.

Can Assist take action in our environment?
Assist reads from Dynatrace to investigate. Action is a governed progression — explain, recommend, prepare, request approval, then trigger a workflow — and what it is permitted to do is configured by you. For sensitive operations, human approval stays in the loop.

Changelog

Changelog will populate here as Assist releases ship. Check back after the next release.