Fair use policy

Last updated: 2026-04-23

Luma BI is a managed platform — your data lives on dedicated Postgres + Redis infrastructure that we provision and operate for you. Fair use exists so both sides have a predictable cost picture and so one customer's usage spike never affects another customer's performance.

This page is the canonical reference for the limits behind every plan tier. Nothing here changes mid-contract; we update it openly with a date stamp when something changes.

What we limit, and why

Four dimensions:

| Dimension | Why we limit it | | --- | --- | | Stored data | Direct cost — your dedicated Postgres scales with row count | | Historical window | Determines how much of your warehouse we hold in your Luma instance | | Sync cadence + connector concurrency | Determines how hard your warehouse and our pipelines work | | AI Analyst spend | Anthropic API consumption — passes through to a real bill |

We deliberately do not limit: number of users, number of dashboards, number of saved AI Analyst reports, query depth on dashboards, or sharing.

Limits by tier

| Limit | Starter | Growth | Enterprise | | --- | --- | --- | --- | | Historical data window | 36 months | 60 months | Unlimited | | Stored data ceiling | 50 GB | 500 GB | Custom (typically 5 TB+) | | Warehouse / API connectors | 1 | Unlimited | Unlimited | | Sync cadence | Daily | Every 1h–24h | Sub-hourly | | AI Analyst monthly spend | $50 | Pooled, budget-capped | Custom | | AI Analyst per-user daily run cap | 2 | 5 (configurable) | Configurable | | query_sql row LIMIT | 10,000 (hard ceiling 50,000) | same | same | | query_sql timeout | 30 seconds | same | same | | sample raw rows | 20 | same | same | | AI Analyst wall-clock per run | 15 minutes | same | same | | AI Analyst tool-call iterations per run | 25 | same | same |

The historical data window

The single most important limit to understand. Each tier holds a rolling window of your data on your dedicated Luma Postgres instance:

  • Starter — last 36 months. Enough for year-over-year seasonality, most cohort analysis, and 2-year vintage curves.
  • Growth — last 60 months. Full 5-year view; covers fintech vintage curves, multi-year retention cohorts, long-tail subscription LTV.
  • Enterprise — unlimited. We hold everything your warehouse holds.

Data older than your window stays in your warehouse / source — we just don't hold a copy. Dashboards and the AI Analyst can only see what's in your Luma instance.

When you upgrade tier, we trigger a one-time backfill from your warehouse to extend the window. This typically completes within 24 hours; we coordinate the timing with you. CSV-only customers can re-upload older exports any time through the admin UI.

Stored data ceiling

The other dimension that affects our infrastructure cost directly. The ceilings are sized to comfortably absorb the typical data volume that fits inside each tier's historical window — they exist to protect against the edge cases (someone with billions of clickstream events on Starter, a 10 TB warehouse on Growth) rather than to gate normal use.

If you approach a ceiling we will:

  1. At 80% — email your account contact with current usage, projected exhaustion date, and the upgrade or trim options.
  2. At 95% — second notice, escalation to a phone call.
  3. At 100% — pause new ingest, but dashboards and the AI Analyst keep working on the data already loaded. We do not auto-suspend the platform.

In practice customers either upgrade tier, agree a custom storage add-on, or trim what they sync (most over-runs are one fat table that doesn't need to be in BI).

Sync cadence and connector concurrency

Each connector runs on a schedule we configure during onboarding within your tier's allowed cadence range. We don't throttle individual syncs — but we do limit how many can run concurrently to keep your warehouse load and our pipeline costs predictable. Burst-sync (e.g. an ad-hoc full re-sync) is always available; just ask.

AI Analyst spend

The AI Analyst calls Anthropic's Claude API, which has a real per-token cost. To keep your bill predictable:

  • A daily dollar cap stops new runs once today's spend hits the cap (default $5/day per admin)
  • A per-user daily run cap stops one user from monopolising the budget (default 2/day for non-admins)
  • Errored runs do not count against either cap — transient rate limits never burn quota
  • Both caps are configurable up to the tier's monthly ceiling

Admins can see live spend on the Admin · AI Spend dashboard.

Acceptable use

Luma BI is licensed for commercial analytics and reporting. The following are not acceptable:

  • Resale or sublicensing of dashboards, reports or AI-generated content to third parties
  • Storing or processing data you don't have rights to (PII without legal basis, scraped third-party content, etc.)
  • Non-analytics workloads via the AI Analyst — the SQL tool is for reporting questions, not for scheduled ETL or operational queries
  • Reverse engineering the platform or extracting our compute functions for use outside Luma BI
  • Automated scraping of dashboards or AI Analyst output via the public URLs

If something here is ambiguous for your use case, ask us — we'd rather agree the boundary than discover it.

What happens if you exceed a limit

The order of escalation, every time:

  1. Notify — automated email to your account contact, with the specific limit and current usage
  2. Talk — your CSM reaches out within one business day to understand what's driving it
  3. Resolve — upgrade tier, custom add-on, sync trim, or whatever fits the situation
  4. Pause — only if we can't reach you, and only the new ingest / new AI runs. Existing dashboards keep working.

We have never auto-suspended a customer.

Changes to this policy

We update limits openly with a date stamp at the top of this page. Material reductions to existing customer limits require 60 days' notice. Increases (more headroom) take effect immediately.

Questions? Email us — we'll answer and add anything new to this page.