WorkerKitAI workers for PostHog
AI workers for PostHog read your product analytics on a schedule and tell you what moved: the funnel step that leaked overnight, the trend that broke, the error that started after a release, the experiment that is ready to call. They run the same queries your team runs in PostHog, on the project you choose, with only the tools you grant.
What is an AI worker for PostHog?
An AI worker for PostHog is an AI worker given one recurring job in one PostHog project: run funnel, trend, retention, paths and SQL queries, read saved insights and dashboards, list session recordings, read error tracking issues, feature flags, experiments, surveys and web analytics, and, where you grant it, create or change those objects. It connects with a personal API key you create in PostHog with the MCP Server preset, so the key decides the project and the reach. Each worker gets only the tools you give it.
What these workers do
Daily funnel drop watch
Every morning, run yesterday's funnel, compare each step with the week before, and when a step drops past your threshold, pull the recordings behind the drop-off and say what those sessions had in common.
Release morning-after report
The day after a release, compare the key trends and the error tracking issues against the previous days and name what changed, with the insight and issue links.
Experiment readiness check
Each weekday, read every running experiment's results and tell the team which ones have enough exposure to call, which are flat, and which have a metric moving the wrong way.
Weekly product usage brief
Once a week, read the saved insights that matter, retention and the top paths, and write the one-page brief a product manager would otherwise assemble by hand.
Survey response digest
Summarize new survey responses into themes and counts, with representative quotes, without anyone exporting a spreadsheet.
How an AI worker runs in PostHog
One key, one project
Reads first, writes on purpose
Ask for comparisons, not totals
Mind the request budget
AI-powered tools are optional
PostHog questions, answered.
Can a worker change our PostHog setup?
Only with the tools you grant it. Running queries and reading insights, recordings, flags and experiments changes nothing. Creating or editing insights and dashboards, changing a feature flag, launching a survey or ending an experiment are separate tools, granted per worker, and they act in PostHog as the user whose key is connected.
Which project does a worker see?
The project the personal API key was created for. The MCP Server preset in PostHog scopes a key to one project, so a worker can never wander into another one. For a second project, create a second key and connect it to its own workers.
Can it watch session recordings?
It can list the recordings behind a funnel step or a filter and read what PostHog recorded about each session, such as the pages, the events and the console logs. That is enough to say what the sessions had in common; it does not play video.
Does it work with self-hosted PostHog?
No. The connection uses PostHog's hosted service for PostHog Cloud projects in the US or EU region. A self-hosted PostHog has no hosted connection to offer.
How much can a worker query?
PostHog limits each key to a number of requests per minute and per hour, and query endpoints carry their own budgets. Every worker sharing a key shares those limits, so give each worker a few targeted queries rather than a sweep, and spread schedules when several workers use one key.