LiteLLM, for Customer Success
For Customer Success Managers
LiteLLM is a self-hosted AI gateway that sits between a customer's applications and many model providers. Apps send one OpenAI-style request; LiteLLM routes it, tracks spend, and controls who can use which models.
Takeaways
Actors, inputs, and outputs
Concepts
Processes and services
Stores
Infrastructure
100%
Your role
Why this matters for a Customer Success Manager
- When a customer says they use LiteLLM, they usually mean a shared gateway, not a Python library inside one app.
- Ask whether apps call vendors directly or go through the gateway. That changes how you diagnose outages, spend spikes, and model-switch requests.
- Spend questions belong on virtual keys, teams, and the admin view, not on each application's logs.
- An Anthropic or OpenAI outage may not be a product incident if fallback is already configured in the gateway.
- SSO, audit logs, and contracted support are LiteLLM Enterprise add-ons. The open-source gateway already covers unified access, virtual keys, budgets, and routing.
How the pieces fit
- Apps send one OpenAI-style request to LiteLLM.
- A virtual key authorizes the call and can carry a budget.
- LiteLLM routes the call to a model provider and records the spend.
Terms customers may use
- Proxy, AI gateway, and LiteLLM often mean the same shared service.
- The Python SDK is a library one app can import. The gateway is what the whole organization calls.
- OpenAI-compatible means apps keep the OpenAI request shape even when the backend is Anthropic, Bedrock, or Azure.
Why customers use it
- Stop integrating every vendor API in every app.
- See and cap AI spend before the invoice arrives.
- Change models without starting a new application project.