Berri AI offers LiteLLM—an open-source AI gateway (LLM proxy) that places the entire AI stack behind a single OpenAI-compatible key. It is aimed at platform teams for centralized access, observability, control, and optimization of calls to various large models, agents, and MCP.
Purpose The core purpose of LiteLLM is to serve as a unified control point for enterprise AI calls: allowing applications to call numerous model providers through one API, while enabling platform teams to centrally perform cost tracking, budget control, routing, caching, and governance. It is officially stated to support 140+ providers and 1,892 models.
Main capabilities Access: provides a single OpenAI-compatible API that can connect to 140+ providers and 1,800+ models, so switching models requires no changes to application code; supports SSO single sign-on and divides permissions by team, project, or application; provides "zero-day support" for new models; in addition to LLMs, agents and MCP servers can also be accessed through the same gateway; supports connecting to proprietary, fine-tuned, and self-hosted models. Visibility and control: tracks usage and spend by key, user, team, organization, tool, agent, and MCP, covering 140+ providers; supports enterprise-grade chargeback billing; allows hard budgets to be set by key, team, organization, and model, reset daily/monthly, with requests stopped once the limit is reached; provides rate limiting and leaked-key protection; provides model access control and guardrails, and retains audit logs for every request. Cost optimization: switch models without changing code; Auto Routing can send simple prompts to cheaper models and difficult prompts to stronger models; supports load balancing across providers, regions, and keys; lowest-cost routing; response and semantic caching (Redis, S3, GCS); prompt compression to reduce token consumption. Deployment: can be self-hosted in your own cloud or in a fully air-gapped environment; provides official Helm charts or Terraform modules; provides official Docker images (including database-bundled and non-root variants); can run on your own Postgres and Redis, and autoscale via Kubernetes; supports one-click deployment to AWS, GCP, or Azure. It is officially stated that its Rust AI Gateway adds only 0.66 ms at p99 in benchmarks.
How to use of the official approach is: deploy the open-source gateway, then track and limit spend on every request; when organization-level use is needed, layer on SSO, audit logs, and SLA. Deployment can use the official Helm chart or Terraform module, run on your own Postgres and Redis, and autoscale with Kubernetes.
Suitable scenarios Suitable for platform teams and enterprises that need unified management of multiple model providers, control of AI spending, cost allocation by team, centralized governance of model access, and have self-hosted or air-gapped deployment requirements.
Limitations and notes The LiteLLM gateway is MIT licensed and can be self-hosted for free; the enterprise edition adds SSO, RBAC, audit logs, SCIM, and dedicated support on top of the same open-source core. It is officially stated that self-hosting does not send telemetry, and prompts are sent only to the configured model providers. Regarding specific pricing, SLA terms, and the complete feature list, the information is not provided and must be checked in the official documentation.
