AI Query Generator

You type a question or description in the search bar, prefixed with a forward slash or by clicking the AI icon. The input can be anything from a simple phrase like "show me errors from the payment service" to a complex request like "compare error rates between us-east-1 and eu-west-1 over the last 24 hours, broken down by service."

The generator supports English and Dutch input. It handles typos, abbreviations, and informal phrasing — you do not need to be precise.

Before generating a query, the AI inspects your environment to understand what is available. It considers the following context:

This context injection is what makes the generator accurate for your specific environment. A generic AI model would not know that your team uses "payment-svc" as the source name for the payment service — but the LogPulse generator does.

The final stage compiles your intent and the environment context into an optimized LPQL query. The generator applies several optimization rules:

1. It places the most selective filters first to maximize partition pruning in ClickHouse.

2. It uses field expressions instead of full-text search when a specific field is known, which is significantly faster.

3. It adds appropriate pipe commands for aggregation, sorting, and limiting when the question implies a summary rather than raw log output.

4. It infers time ranges from your input (for example, "yesterday" maps to earliest=-1d@d latest=@d) and combines them with the currently selected range.

The fastest way to use the generator is directly from the search bar. Click the sparkle icon (or press /) and type your question. The generated LPQL query appears below the input as a preview. Press Enter to execute it, or Tab to insert it into the search bar for further editing.

> show me all 5xx errors from api-gateway in the last hour

Generated LPQL:
source=api-gateway level=error attributes.status_code>=500 earliest=-1h
| stats count by attributes.status_code
| sort -count

For more complex queries, open the AI panel by clicking "AI Query Generator" in the toolbar or pressing Ctrl+K (Cmd+K on macOS). The panel provides a larger input area and shows a step-by-step breakdown of how the query was constructed.

In advanced mode you can also:

1. Pin context fields — force the generator to always include certain filters (for example, always scope to your team's namespace).

2. Set output preferences — tell the generator whether you want raw logs, a count summary, a timechart, or a table.

3. Chain follow-up questions — ask "now break that down by region" and the generator modifies the previous query instead of starting from scratch.

After a query is generated, you can refine it in two ways:

Edit the LPQL directly — the generated query is plain text in the search bar. You can adjust filters, add pipe commands, or change the time range manually.

Ask a follow-up — type another natural language question and the generator will modify the existing query rather than replacing it. For example, after generating a query that shows errors, you could say "also include warnings" and the generator will update the level filter from level=error to (level=error OR level=warn).