What you can analyze
Almost any analysis you can phrase is possible: these guides are starting points, not a catalog.
The guides in this section are worked examples, not an exhaustive list. Because the assistant reasons over the full range of metrics Datasentinel collects, almost every type of analysis you can phrase becomes a use case. Pick the guide closest to your question and adapt it to your own instances and scenarios.
Browse the guides by goal
Start here
Navigate your fleet and platforms
Tune performance
Troubleshoot an instance
Compare and track change
Scale and automate
Analysis areas
The main analysis areas the assistant covers:
Workload and sessions
Session load over a window, activity spikes, who was active and when
Queries
Top queries, single-query deep-dive with statistics and execution plan, live plan of a statement
I/O, WAL, and checkpoints
I/O-bound windows, WAL volume, checkpoint pressure, slow commits
Tables and indexes
Bloat, sequential scans, unused or missing indexes
Locks and blocking
Blocking chains, root blockers, contended objects, deadlocks
Replication
Standby lag, WAL backlog, paused replay
Configuration and system
Parameter changes and drift, host CPU, memory, swap, load
Live state
Current sessions, live blocking, VACUUM and index-build progress
Comparisons and diagnosis
Two time windows or workloads side by side, broad triage when you don't know what's wrong
Each area also has a named command, such as datasentinel ash or datasentinel locks, for when you already know which analysis you want. The full list, with what each command needs, is in the Skill commands reference.
Keep, export, and revisit your analyses
An analysis is not lost once the conversation moves on. AI Connector takes full advantage of the capabilities of the assistant it runs in:
Export as a file. Every analysis can be exported as a self-contained interactive HTML report or as a portable Markdown file, ready to share or to paste into a ticket, a wiki page, or a pull request. The platform can also generate PDF reports server-side. See Generate and share reports for the details of each kind and example prompts.
Come back to it later. The analysis stays in the assistant's conversation history, so you can reopen the session a few days later, re-read the findings, and continue the investigation from where you left off.
Dynamic charts, where your assistant can draw them
Analyses do not have to stay text-only: the assistant can draw dynamic charts of your metrics alongside its findings.
How far that goes depends on the assistant you use and its type:
Desktop apps and IDEs can typically render interactive charts inline with the analysis. Some go further: Claude, for example, can turn an analysis into a full interactive dashboard, as shown in Claude Artifact.
CLI assistants work in a terminal, so charts render as text or not at all. There the assistant presents the same data as equivalent Markdown tables.
Whatever the rendering, the analysis itself is identical: the charts and tables are two views of the same findings, and the exported HTML report keeps its interactive charts in every case.
On large fleets, tag your instances
Cross-instance analyses select their instances by tag (TAG): every production server, all instances of an application, a whole datacenter. On a large fleet, the quality of those analyses therefore depends directly on the quality of your tagging.
See Analyze across multiple instances for cross-instance example prompts.
Last updated















