AI and your data

What gets sent to the model?

HitLight sends the current request and the minimum show context needed to answer it to OpenAI. Fixture-profile generation sends the manual, image, or pasted channel table you explicitly provide.

What are the usage limits?

During the public beta, verified accounts receive 200 assistant turns per month by default, limited to 50 per hour and 20 model calls per turn. These limits are enforced server-side and may change while we tune capacity. Requests are routed through our own server, so you never supply or store an OpenAI key.

How long is any of it kept?

Raw per-turn operational diagnostics expire after 30 days; submitted feedback expires after 90 days. Both are also removed with the show or the account. Uploaded fixture sources remain private to show collaborators and authorised reviewers. They are not analytics properties.

What does AI observability collect?

Optional AI observability is disabled until the person submitting the turn consents. An opted-in PostHog trace contains model, timing, token, cost, outcome, and tool-name metadata only. It never contains prompts, replies, shared history, system instructions, tool arguments or results, show state, DMX values, Auto Show vibe or palette text, fixture manuals, filenames, or generated fixture definitions.

Product analytics identify a signed-in person only by their internal Firebase user ID — never by name or email. A remote user's AI turn follows that user's consent and identity rather than the master device's settings.

Can the AI touch dangerous channels?

No. AI never automatically arms flame, pyro, ignition, or similar hazardous controls. A generated fixture definition stays a private draft until its channels are tested and confirmed by an operator. The full exclusion list is on lighting safety.

Questions or deletion requests: privacy@hitlight.app.

Last verified 1 August 2026.