| Hard budgets (team / key / user) (scored) | yes Budgets per virtual key, team, team member, internal user, customer and agent; multiple budget windows; optional fail-closed enforcement. Requires a Postgres database; per-model budgets need Enterprise Source: docs.litellm.ai/docs/proxy/users |
|---|---|
| Rate & spend limits (scored) | yes RPM/TPM and max parallel requests per key, user, team, model and agent Source: docs.litellm.ai/docs/proxy/users |
| Cost attribution / chargeback tags (scored) | yes Spend by key, user, team and org; tag-based spend tracking Source: litellm.ai/enterprise |
| FOCUS export (scored) | yes version: 1.2 Experimental export conforming to FOCUS 1.2 (Parquet/CSV to S3, GCS, Vantage; also Ternary and Mavvrik connectors) |
| EU data residency option (scored) | yes Self-hosted: runs in any region you choose Source: litellm.ai/enterprise |
| Public pricing (scored) | yes Source: litellm.ai/enterprise |
| Open source | yes MIT for the core; the enterprise/ directory is under a commercial license |
| Hosting | Self-host Self-hosted or air-gapped in your own infrastructure Source: litellm.ai/enterprise |
| Supported providers / models | 100+ "Call 100+ LLM APIs" Source: github.com/BerriAI/litellm |
| Caching | yes Redis/Valkey exact cache and Redis semantic cache |
| Pricing model | Open source + Enterprise license Source: litellm.ai/enterprise |
| Public price points | OSS free; Enterprise: custom pricing (talk to sales), 30-day trial key Source: litellm.ai/enterprise |
"unknown" means we could not confirm it in public sources. Vendors can send evidence to update a cell; see get verified. Verification never changes the scoring rules.