Bifrost
Every LLM request, attributed to a team.
Bifrost by Maxim AI estimates the cost of every LLM call through your gateway. Our open-source exporter sends it to Cloud Intelligence™ by provider, model, virtual key, team and customer, right next to your cloud bill.
Open-source exporter · No prompt content · Visible in about 15 minutes
Bifrost · October
what we import
Bifrost data, in the dimensions you already use.
Each gateway request maps to Cloud Analytics provider, service and SKU, plus Bifrost labels for model, virtual key, team, customer, feature, business unit and project. Costs are estimates for showback.
Bifrostanthropicclaude-sonnet-4-20250514what you can do with it
From gateway logs to a decision
Cost by virtual key
Slice estimated spend by virtual key, team, customer, feature or model.
Error rate
Failed requests export at zero cost, so error rates stay reportable.
Input vs output cost
Input and output cost, side by side, on gateway 2.0.0 and later.
GenAI Intelligence
Bifrost next to your other AI providers in one dashboard.
Budgets and alerts
Guardrails on gateway spend before provider invoices land.
Allocations
Put estimated LLM spend on teams and customers, from Bifrost labels.
live in minutes
Connect Bifrost in three steps
Create an API token.
In your Cloud Intelligence™ profile, with the DataHubAdmin scope. Shown only once.
Run the exporter.
One container beside your gateway, via Docker, Helm, docker-compose or a binary.
Reports in minutes.
Run once to verify. Data appears in Cloud Analytics in about 15 minutes.
Open source, Apache-2.0. The exporter never reads prompt or response content.
Start freeFrequently asked
questions
How does the Bifrost integration connect?
You run the open-source bifrost-datahub-exporter next to your self-hosted gateway. It polls the gateway's logs API and pushes events to our DataHub API, so we need no network access into your environment and your gateway credentials stay in your network.
What do I need to get started?
A Bifrost gateway version 2.0.0 or later with logging enabled, a DataHub subscription included in your tier, and an API token with the DataHubAdmin scope. Older gateways work best-effort without the input and output cost split.
Is this my billed LLM cost?
No. Bifrost bills nothing, so these are estimates from its model-pricing catalog for attribution and showback. Your authoritative spend still arrives through your provider and cloud billing feeds, so don't add the Bifrost dataset to bill totals.
How fresh is the data, and can I import history?
The exporter polls every five minutes by default, and events appear in Cloud Analytics about 15 minutes after they are sent. Set BACKFILL_DAYS on first run to import history, bounded by your Bifrost log retention and the DataHub two-year window.
Which dimensions and labels are available?
Service is the upstream provider and SKU is the model. Labels add virtual key, team, customer and feature, plus business unit and project when your Bifrost plan populates them. Failed requests export at zero cost, tagged as errors.
How does Bifrost show next to my other providers?
Bifrost appears as its own provider in Cloud Analytics and in the GenAI Intelligence dashboard alongside other supported GenAI providers. Every event carries a cost_basis: estimated label, so you can keep it separate from billed spend.
Connect Bifrost. Know who drives every LLM call.
Open-source exporter. No prompt content. Data in minutes.