My Dynatrace Assist Tips and Tricks

[andreas_grabner]

Hi

Many of you have already played around with Dynatrace Assist. So have I. And I have just published a short Tips & Tricks video to walk through several use cases on YouTube.

In the video I also try to beat Dynatrace Assist in analyzing traces. Well - I ended up in 2nd place 🙂

Please share your own assist conversations. What works, what doesnt, what surprises?

[txvepr]

Just recently I had the Dynatrace AI create an entire dashboard based on approximately 12 rows of logs that I had from a test on 2 separate days. I needed to be able to have this Dashboard operating when we go live, but only had those 2 days of short testing from the vendor making the Posts. I told it my instructions and gave it the 2 dates to check. 

Based on those 12 rows of test posts having a variety of elements tested, It created a 12 pane dashboard including Geolocation map of those posts and it is ready for Go Live tracking.  

Simply Amazing!

[andreas_grabner]

cool. thanks for sharing. Any screenshots you can share? Or any parts of the prompt? Its always great to learn from others!

[txvepr]

Creating a pipeline and Dashboard from minimal log entries in Dynatrace:

Background:

This POC needed to create an Open Telemetry Pipeline for logs to a dedicated bucket for audit purposes.

I only had 2 log files with only 12 rows of entries from a quick test to work with . Of course, once we get closer to go live we will get more. The logs are in a JSON format.  Investigate your log first to look for patterns or meaningful elements.

1. Based on your DQL from above, Dynatrace AI query then can be : “how can the logs returned from this DQL be a pattern for pipeline detection: fetch logs | filter contains(content, "xxxx", caseSensitive: false) // Add filters by: | filter matchesValue(field, "value") “
2. Since there are multiple values possible instead of just “xxxx” , I then prompted “not based on xxxx content”
3. DT AI then went back and discovered a better pattern match for this particular Open Telementry feed ingested log to use for identification.
4. I queried “can a pipeline be based on the other data elements like server, hostname, ip or other non content related attributes”  . “Yes, absolutely”
5. DT AI then listed out that host.name would be the best identifier . It returned DQL for me to run to find what non-content fields are populated.
6. I then asked if “data.audience” could be used since it will be unique to this particular feed because the host or IP could change in the future.
7. Just to verify correctness I then told DT AI to use my 2 logs I had from their dates rather than just based on my DQL at the start. “use example data from the date 2026-06-23T00:00:00-05:00/2026-06-23T23:59:59-05:00 “
8. DT AI returned a “Fields produced by the Pipeline” in a table with the validation DQL.
9.  DT AI then asked if “Would you like to add alerting on xxxx.auth.failed == true, or build a dashboard summarizing activity by xxxx.api.endpoint and xxxx.organization?*”
10. I chose “build a Dashboard”

It gave me a 12 panel Dashboard that works plus the import JSON.

Obviously you should always verify the output with what you know about your own data before accepting it. 

If you haven’t started leveraging the DT AI, you are missing out on a huge force multiplier for your team.

[andreas_grabner · Accepted solution]

Thanks a lot for sharing the details.

One final question: Did you do this within Dynatrace in Dynatrace Assist? Or did you use a Claude/CoPilot/... and connected it to Dynatrace via MCP or dtctl?

[txvepr]

This was done entirely within the Dynatrace Assist. The ways it can correlate across our stack is what makes this feature indispensable.

[andreas_grabner]

Thanks. FYI -> @GabrieleHB 

[NicoleMT]

Hi Andy!!, I have a question. Does using the new Dynatrace Assist features incur additional subscription costs or consume subscription credits? If so, could you please let me know which features or use cases generate those costs?

[Julius_Loman]

@NicoleMT Assist itself does not have any costs associated. Only the DQL assist executes incur costs based on your rate card. Assist skills have limits in place.

[Frank_Madeja]

Hi,

I’ve been using Dynatrace Assist for a few weeks now. In many cases it works very well, but sometimes even very simple questions can’t be answered. For example, I asked:

“How many OneAgents are currently running in discovery mode, infrastructure mode, and full‑stack mode?”

After three rounds of “continue investigation” and spending a fair amount of token budget, it still couldn’t provide a result.

To be clear, I think Dynatrace Assist is a great feature, but it would be beneficial for all of us if its ability to answer environment‑specific, data‑driven questions like this could be further improved. Thanks!

[andreas_grabner]

Servus Frank 🙂

Thanks a lot for your feedback. Thats very valuable and I will forward this to our Assist PMs - sounds like we need to update some of the skills. 

Just have one quick question: The Dynatrace Assist can run in "Chat Mode" and in "Agentic Mode". To answer your question it needs to run in Agentic Mode as in Chat Mode it is not allowed to make any queries on your behalf. Just wanted to double check with you if you have executed the query in Agentic Mode.

To enable agentic you need to go to your settings. Plus - when you open the Assist in the UI on the bottom left you also see if you are in Agentic mode or not. If you open the Assist from an "Explain ..." button on the UI then you are in Chat Mode. To get into Agentic simply create a new conversation with the + button

Andi

[andreas_grabner]

Actually Frank - I can easily reproduce this behavior. Its clearly a Skill issue. I will forwadr this to the PM

[txvepr]

Frank, 

When I ran your exact query, i got the results correctly for our system . Its set to agentic AI also.  Are you on a SAAS or self managed system?

OneAgent Monitoring Mode Distribution

Monitoring Mode Host Count

Full-Stack	14
Discovery / Infrastructure / Unknown	2
Total Hosts	16

---

Notes

• 14 hosts are running in Full-Stack mode — these receive complete observability including application performance monitoring, code-level visibility, and distributed tracing.
• 2 hosts have a null monitoring mode, which typically indicates they are in Discovery mode (OneAgent installed but not yet assigned a monitoring mode) or Infrastructure mode (host-level metrics only, no deep process injection). A null value can also appear briefly during agent initialization.
• No hosts were explicitly found with INFRASTRUCTURE or DISCOVERY mode labels in the entity data.