Tools
LiteLLM moves prompt-injection heuristics off the event loop and starts 429ing saturated deployments
LiteLLM v1.103.0-rc.1 (2026-09-20) includes #41541, which runs proxy prompt injection heuristics off the asyncio event loop — a guardrail that was blocking the loop for every request it inspected — and #41555, which rejects with 429 when a deployment's `max_parallel_requests` slots are all in use instead of queueing. Other operationally relevant items: JWT key-mapping cache eviction on user, team, org and bulk key deletion (#41707), config file now winning over the database (#41779), and `save_config` persisting only the keys a caller changed (#41748). A OpenRouter price sync in the same release moved 443 models with 191 new and 4 deprecated.
Source
↳ Follow the thread