Investigate API Performance Bottlenecks

Zero pulls TP95 data from Axiom, surfaces the slowest endpoints, identifies patterns, and creates a GitHub issue with findings from one Slack message.

Zero connects:

What Zero delivers

Investigate API Performance Bottlenecks — sample output from Zero — screenshot 1

What the problem is

TP95 spikes don't always page you. They sit in your metrics dashboards, quietly degrading user experience while your team is focused on shipping. Zero makes it easy to check any endpoint's performance on demand - pull the last 7 days of TP95 data, see which requests are slow, and get a GitHub issue filed automatically so nothing gets dropped.

How Zero fixes it

Step 1: Connect your tools

Axiom

Required

Axiom is required. Zero queries your Axiom dataset for request events filtered by endpoint path and time window.

Connect

GitHub

Optional

GitHub is optional. Zero creates issues from findings when asked, with the metric table embedded in the body.

Connect

Step 2: Ask Zero

@Zero check Axiom for the POST /api/zero/runs endpoint - show me the TP95 for the last 7 days and flag any events over 5s.

try it

Zero queries Axiom for the endpoint

Zero pulls all request events for the specified endpoint over the time window, computing TP50, TP95, and TP99 per day.

Zero surfaces slow events

Zero filters for events above the threshold (default: 5 seconds), groups them by time of day and error pattern, and identifies the likely root cause.

GitHub issue created with findings

Zero files a structured GitHub issue with the metric table, slow event list, and root cause analysis - ready for the engineering team to act on.

Step 3: Take it further

Assign and prioritize

Route the issue to the right engineer

@Zero assign the Axiom issue to Ethan and add the label performance, priority-high.

try it

Schedule regular monitoring

Set up a weekly TP95 check for key endpoints

@Zero every Monday at 9am, pull the last 7 days of TP95 for /api/zero/runs and post to #dev. File a GitHub issue only if TP95 exceeds 3s.

try it

Investigate a specific slow event

Dig into a single outlier event

@Zero pull the full trace for the 6,857ms event on Apr 12 from Axiom and tell me where the time is spent.

try it

Tips for better results

●Specify a threshold in your prompt - 'flag events over 3 seconds' - so Zero's findings match your SLO, not a generic cutoff.

●Ask Zero to check the endpoint right after a deploy to catch regressions before they affect users at scale.

●Combine with Daily Error Triage for a complete morning health check: errors from Sentry plus performance from Axiom in one brief.