LiteLLM
LiteLLM spend, beside your cloud bill.
Cloud Intelligence™ brings in the spend your self-hosted LiteLLM proxy meters on every call, labeled by provider, model, virtual key, team, feature and end customer. Right next to the rest of your AI and cloud spend.
Open-source exporter · Never reads prompts or responses · Polls every 5 minutes
LiteLLM · Octoberwhat we import
LiteLLM data, in the dimensions you already use.
Every model call maps to standard Cloud Analytics dimensions, plus labels for provider, model, virtual key, team, feature, customer and tag.
LiteLLManthropicclaude-sonnet-5what you can do with it
From gateway logs to a decision
Cost by feature
Slice by feature, customer, virtual key, team or tag.
Feature tag coverage
See how much spend carries a feature label.
Provider vs provider
Tokens and estimated cost per upstream provider, side by side.
GenAI Intelligence
LiteLLM next to your other AI providers in one dashboard.
Budgets and alerts
Guardrails on gateway spend before the provider invoices land.
Allocations
Put LiteLLM spend on teams and customers, by key or label.
live in minutes
Connect LiteLLM in three steps
Create an API token.
In your profile, with the DataHubAdmin scope. It's shown only once.
Run the exporter.
One small container beside your proxy. Your LiteLLM admin key stays local.
Spend in minutes.
Events appear in Cloud Analytics about 15 minutes after export.
Open source, Apache-2.0. Stop the exporter any time.
Start freeFrequently asked
questions
How does the LiteLLM integration work?
You run the open-source litellm-datahub-exporter next to your proxy. It polls your proxy's spend APIs and pushes events to the DataHub API, so we need no network access into your environment.
What do I need before I start?
A self-hosted LiteLLM proxy, version 1.65 or later, with spend logging enabled, a DataHub subscription in your tier, and an API token with the DataHubAdmin scope. On v1.65-era proxies, run the exporter in daily mode.
Is the cost an invoice figure?
No. LiteLLM estimates cost from its model price map and reported tokens, so treat it as showback. Every event carries a cost_basis: estimated label, and you can reconcile with our Anthropic and OpenAI integrations.
How fresh is the data, and how much history can I import?
The exporter polls every five minutes by default, and events show up about 15 minutes after they are sent. Set BACKFILL_DAYS on first run; history is bounded by your spend-log retention and DataHub's two-year window.
How do I attribute spend to features and customers?
Set the OpenAI user parameter for the customer and pass a feature in spend_logs_metadata. Team and virtual key come from the LiteLLM key, and tags come from LiteLLM request tags.
Does the exporter see my prompts?
No. The exporter never reads or transmits prompt or response content, and your LiteLLM admin key never leaves your network.
Connect LiteLLM. Know what every model call costs.
Open source exporter. Spend in minutes.