Frequently Asked Questions

FinOps tools track cloud infrastructure spend at the resource level (VMs, storage, GPU hours). Bear Lumen goes further: we reconcile AI costs with revenue data, produce margin reports by customer and feature, and tell you when pricing needs to change. A FinOps tool shows your total OpenAI spend. We tell you that Customer X on Feature Y is underwater, and recommend what to adjust.

Billing tools handle the transaction. Bear Lumen handles what they don't: reconciling provider costs with revenue, producing margin reports by customer, feature, and model, and letting you model pricing changes with a simulator. Keep your billing system. We plug into it and do the cost-and-margin analysis.

Teams that need to see AI cost and margin clearly. If you build AI products, Bear Lumen produces margin reports by customer and feature, flags underwater accounts, and lets you model pricing changes before you make them. If you use AI internally, it produces per-team cost reports and tracks spend against your budgets.

Nobody can see AI cost and margin clearly. Provider invoices land, spreadsheets get updated quarterly (if at all), and by the time you've pieced it together the window to act has closed. Bear Lumen does that work: it reconciles costs with revenue continuously, attributes every dollar to a customer, feature, and model, and lets you model changes before you make them.

Spreadsheets work until they don't. Manually reconciling provider invoices, Stripe data, and usage logs takes days per month and makes scenario modeling painful. Bear Lumen does that reconciliation continuously and gives you a simulator for the scenario modeling. You get the report, not the spreadsheet.

No. Cost tracking works from day one with no customers or billing required. When revenue flows, the same data powers margin analysis and pricing validation.

The SDK captures request metadata (customer ID, feature, model) alongside cost data. We aggregate provider costs and attribute them to individual requests, then produce margin reports by customer, feature, and model.

No. Bear Lumen plugs into your existing billing system (Stripe, Orb, etc.) to get the revenue side. We connect that with provider cost data and handle the margin analysis, reporting, and recommendations. Nothing changes about how you invoice.

Flat rate, volume pricing, tiered/graduated pricing, and hybrid (base fee + usage). The pricing scenario modeling lets you model any combination against historical usage data.

Adding the SDK to your codebase takes minutes. Full setup (connecting revenue sources, defining dimensions) depends on your stack. First pricing decisions typically arrive within the first week. Early customers get direct onboarding support.

Yes. The platform includes demo data so you can see how cost attribution, margin analysis, and pricing recommendations work before connecting your own providers and revenue sources.

No. The SDK reads the response your provider already returned, so nothing sits between you and the model. Tracking events queue in memory and flush in background batches, the same pattern analytics SDKs have used for years.

We collect request metadata and cost data. All data is encrypted in transit (TLS 1.3) and at rest (AES-256). No raw prompts or customer PII. You own your data and can export or delete it anytime. Building toward SOC 2 Type II certification.

Bear Lumen offers a free Developer plan with 10K events per month. Growth ($49/mo) includes 250K events and Scale ($199/mo) includes 2M events. Usage beyond your plan limit is billed per event at your plan’s overage rate. Enterprise pricing is custom. See the pricing page for the full breakdown.

The system works at any scale. Smaller customer bases mean faster integration and clearer data. Many teams find it easier to establish cost tracking early.

A billed event is a single tracked AI call you send through the SDK or API: one chat completion, one embedding request, one tool call. We read the token counts, model, cost, and the metadata you attach. Your monthly limit counts these events, not tokens, and we never store raw prompts or completions.

Yes, on Enterprise plans. Self-hosted and custom models are supported alongside dedicated ingestion queues, a data processing agreement (DPA), multi-year retention, and a custom SLA. Talk to the founder and we will shape a plan around your stack.

Attach your own user or account ID to each usage event. Bear Lumen computes the cost of every model call from its token usage and rate, then groups those costs by the ID you sent, giving you cost to serve and margin for each customer.

No. Bear Lumen records token counts, model, cost, and the attribution metadata you send. Prompt and completion content is never stored.

OpenAI, Anthropic, Amazon Bedrock, Google Gemini, Mistral, and Ollama, with automatic provider and model detection from the response.

It depends on your stage and pricing, but a common working line is above 40 percent margin per customer as healthy, 20 to 40 percent as at risk, and below 20 percent as unhealthy. The point of tracking unit economics is to know which of your customers fall into each band.

Add the variable costs of serving one customer: model and API spend attributed to them, compute, and any per-customer overhead, divided per customer. The model-spend part requires attributing each request cost to the customer that triggered it.

Cost tracking measures what you spend. Unit economics puts that spend next to revenue to measure profit per customer or per request. You need accurate cost attribution first, then unit economics is the margin math on top.